Download - Asset Allocation in Bankruptcy - NBER
NBER WORKING PAPER SERIES
ASSET ALLOCATION IN BANKRUPTCY
Shai BernsteinEmanuele Colonnelli
Ben Iverson
Working Paper 23305http://www.nber.org/papers/w23305
NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue
Cambridge, MA 02138March 2017
The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.
© 2017 by Shai Bernstein, Emanuele Colonnelli, and Ben Iverson. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.
Asset Allocation in BankruptcyShai Bernstein, Emanuele Colonnelli, and Ben IversonNBER Working Paper No. 23305March 2017JEL No. G3,G33
ABSTRACT
This paper investigates the consequences of liquidation and reorganization on the allocation and subsequent utilization of assets in bankruptcy. Using the random assignment of judges to bankruptcy cases as a natural experiment that forces some firms into liquidation, we find that the long-run utilization of assets of liquidated firms is lower relative to assets of reorganized firms. These effects are concentrated in thin markets with few potential users, and in areas with low access to finance. The results highlight the importance of local search frictions and financial frictions in affecting the allocation of assets in bankruptcy.
Shai BernsteinStanford Graduate School of Business655 Knight WayStanford, CA 94305and [email protected]
Emanuele ColonnelliStanford UniversityDepartment of Economics579 Serra MallStanford, CA [email protected]
Ben IversonNorthwestern University2001 Sheridan RoadEvanston, IL [email protected]
Declining industries, insolvency and distressed firms are unavoidable consequences of an evolv-
ing economy. The ability of an economy to subsequently direct assets to better uses has important
implications for productivity and the speed of recovery following adverse economic shocks (Eisfeldt
and Rampini (2006); Hsieh and Klenow (2009); Bartelsman et al. (2013)). Since economies rely on
courts to resolve insolvency, bankruptcy institutions play an important role in allocating the assets
of distressed firms. Two approaches characterize bankruptcy institutions: liquidation and reorga-
nization (Hart (2000); Strömberg (2000); Djankov et al. (2008)). While the liquidation procedure
winds down the firm and puts all assets back on the market, reorganization aims at rehabilitating
the company whenever possible.
Despite the importance of the bankruptcy system, empirical evidence on key questions is scarce:
How do bankruptcy regimes affect asset allocation and utilization? Are assets in liquidation utilized
similarly to assets in reorganization? If not, what frictions lead to different consequences of the two
bankruptcy approaches?
Theoretically, with frictionless markets, the outcomes of both bankruptcy approaches should be
similar, as both regimes should effectively allocate assets to their best use. This null hypothesis
may no longer hold, however, when frictions are present. For example, conflicts of interests between
claimholders, information asymmetry, and coordination costs in reorganization may lead to ineffi-
cient continuation affecting asset allocation (Baird (1986); Gertner and Scharfstein (1991); Aghion
et al. (1992); Ivashina et al. (2015)). In liquidation, assets may not reallocate to best uses if they
are specific to the firm, and markets are thin with few potential users (Williamson (1988); Gavazza
(2011)). Misallocation may be further exacerbated if potential users of the assets are financially
constrained (Shleifer and Vishny (1992)).
To answer these questions, it is necessary to tackle two important issues. First, there is little
information on how assets are reallocated between firms and how assets are subsequently utilized,
particularly in bankruptcy, when plants are shut down and firms are dissolved. Second, distressed
firms that go through liquidation may be fundamentally different from firms that are reorganized.
This is a common limitation to papers that explore the implications of different bankruptcy codes.
Any comparison between two insolvent firms that experience different bankruptcy regimes may be
biased due to unobserved differences in firm prospects and other characteristics.
In this paper we focus on the U.S. bankruptcy system and compare the consequences of liq-
uidation (under Chapter 7 of the bankruptcy code) with reorganization (under Chapter 11 of the
bankruptcy code) on asset allocation and utilization. To do so, we focus on the real estate assets
used by bankrupt firms, and construct a novel dataset that tracks the allocation and utilization
1
of these assets over time. Real estate assets represent a significant portion of firms’ total capital.1
Moreover, these assets are likely to be highly specific, as the optimal user varies significantly with
building features and location characteristics. For example, an industrial warehouse is unlikely to be
suitable for a retail store, and a restaurant is unlikely to be replaced with a hotel. Further, locations
benefit firms differently as they provide access to customers and suppliers, local labor markets, and
knowledge spillovers (Ellison et al. (2010)).
We combine the U.S. Census Bureau’s Longitudinal Business Database (LBD) with bankruptcy
filings from LexisNexis Law, to obtain a dataset with rich information on 129,000 establishments
belonging to 28,000 bankrupt firms employing close to 4.7 million workers at the time of bankruptcy.
The comprehensive nature of these data allows us to examine the population of bankrupt firms in
the U.S., including small and private businesses. An important methodological contribution of this
work is the creation of geographic linkages that track occupier identities and economic activity at
real estate assets over time. This allows us to capture the allocation and utilization of assets even
when plants shut down and the real estate is vacant, or when it is used for a different purpose than
the original plant.2
To explore long-run (five-year) allocation and utilization of these assets we rely on several mea-
sures. We explore the length of time a location continues to be operated by the bankrupt firm and,
if it does not continue, whether it is occupied by a new firm or remains vacant. In addition to the
occupancy rate, we explore the average number of employees at a given location over time. While
the former measure captures only whether economic activity takes place in a given asset, the latter
measure also captures the intensity of such economic activity.
Tracking assets in bankruptcy reveals several interesting stylized facts. First, both liquidation
and reorganization lead to substantial asset reallocation. Second, when an asset is redeployed to
a different user, it is most likely to a local firm and typically remains within the same industry,
suggesting a significant degree of asset specificity and search costs, consistent with Williamson
(1988) and Ramey and Shapiro (2001). Finally, we find that industry conditions and, especially,
local economic activity are important determinants of asset reallocation and utilization, consistent
with the importance of market liquidity and economic conditions for asset redeployment (Shleifer
1Based on Flow-of-Funds tables from the Federal Reserve, nonresidential structures (value of buildings, excludingthe value of the land) accounted for $8.2 trillion of real assets, while nonresidential equipment comprised only $4trillion at the end of 2014.
2These circumstances are not fully captured by the standard LBD linkages that link plants over time. For example,if an auto parts manufacturer, AutoABC, is shut down, and the building is then occupied by a shoes manufacturer,ShoesXYZ, linkages at the LBD will consider the death of AutoABC and the birth of ShoesXYZ as two separateincidents. Our linkages will connect the two, showing that ShoesXYZ replaced AutoABC in this real estate location.For details on how LBD linkages are constructed, see Jarmin and Miranda (2002). We describe our linkages in detailin Section IV.A and in the Appendix.
2
and Vishny (1992); Gavazza (2011)).
In the main analysis, in order to deal with the endogeneity of the bankruptcy regime, we employ
an instrumental variables strategy that exploits the fact that U.S. bankruptcy courts use a blind
rotation system to assign cases to judges, effectively randomizing filers to judges within each court
division. While there are uniform criteria by which a judge may convert a case from Chapter 11 to
Chapter 7, there is a significant variation in the interpretation of these criteria across judges.
Our empirical strategy compares bankrupt firms that are reorganized within Chapter 11 to firms
that file for Chapter 11 but are converted to Chapter 7 liquidation due to the assignment of the
judge. In effect, otherwise identical filers are randomly placed in either reorganization or liquidation
due to the random judge assignment, thereby allowing us to compare asset outcomes across the two
regimes. Our empirical strategy follows a growing set of papers that takes advantage of the random
assignment of judges and variations in judge interpretation of the law (Kling (2006); Doyle Jr (2007);
Chang and Schoar (2013); Dobbie and Song (2015); Galasso and Schankerman (2015)).
This empirical strategy allows us to explore the following question: if a given firm had not been
reorganized, how would its assets have been redeployed through liquidation?3 We first show that, as
expected, bankrupt plants in liquidation are more likely to be shut down, relative to reorganization,
and they shut down more quickly. But interestingly, even after accounting for the subsequent
reallocation of real estate to new users, liquidated plants are 17.4% less likely to be occupied five
years after the bankruptcy filing, suggesting that in liquidation, on average, assets are less utilized.
In addition, the average number of employees at liquidated locations is significantly lower relative
to reorganized locations. These findings illustrate that bankruptcy regimes have large effects on
long-run asset allocation and utilization.
To better understand which frictions lead to the gap in utilization between reorganization and
liquidation, we explore the role of search frictions in limiting asset allocation. Search frictions
can arise in thin markets with few potential users (Williamson (1988); Gavazza (2011)), and when
potential asset users are financially constrained (Shleifer and Vishny (1992)). Empirically, we rely
on two measures of search costs. First, we create a measure of market thickness which assesses
the extent to which potential users of the bankrupt plant’s real estate reside locally. Second, since
assets typically reallocate to new and local businesses, we explore measures that identify markets
with low access to small business finance.
We find that the drop in utilization is significantly larger in thin markets and areas with low
3We use the terms “reorganization” and “liquidation” to refer to bankruptcy procedures similar to Chapter 11 andChapter 7, respectively. Importantly, this usage of the terms “reorganization” and “liquidation” is separate from theultimate outcome of the bankruptcy. Firms in a reorganization bankruptcy regime can be liquidated if that is theoutcome of the bargaining process. The key difference is that liquidation is forced under a cash auction system likeChapter 7, while it is not with structured bargaining.
3
access to capital. Five years following the bankruptcy filing, plants in thick markets are equally
likely to be occupied regardless of the bankruptcy regime, due to significant asset reallocation
to new users in liquidation. In sharp contrast, liquidated plants in thin markets are over 30%
less likely to be occupied than otherwise-identical assets in reorganized firms. Similarly, we find
no long-term differences in employment across the two bankruptcy regimes in thick markets, but
significantly lower employment when assets are liquidated in thin markets. We also find that local
access to finance affects asset allocation in bankruptcy. In regions with high access to finance, we
find similar levels of utilization for both liquidated and reorganized establishments. But in markets
with low access to finance, liquidated assets are less likely to be occupied and have significantly
lower employment relative to plants in reorganization.4
The results thus far highlight that search frictions and financial frictions can importantly affect
the allocation of assets in bankruptcy. A natural next question is whether the results are driven
by inefficient liquidation, in which high search frictions lead to under-utilization of the assets, or
alternatively, if agency costs in the reorganization process lead to inefficient continuation of the firm
and over-utilization of the assets. Clearly, answering this question is difficult because it depends on
the unobserved opportunity cost of the real estate asset, and thus we do not fully rule out either
possibility.5
While answering this question conclusively is challenging, several pieces of evidence are consistent
with the interpretation that inefficient liquidation leads to under-utilization of assets in thin markets
or markets with low access to finance. First, it is important to highlight that because of the random
assignment, the opportunity cost of the asset is likely orthogonal to judge characteristics, and
therefore our results compare assets with similar opportunity costs, and yet, liquidation leads to
lower utilization, and therefore possible under-utilization. Second, we proxy for the unobserved
opportunity cost by measuring the local vacancy rate of non-bankrupt assets, with the assumption
that areas with low local vacancy rates have high demand for real estate assets and thus high
opportunity costs of vacancy. We find that the effects of liquidation are similar in areas with high
and low local vacancy rates. That is, liquidation leads to lower utilization when search costs are
high, even if the opportunity cost of doing so is high.
In addition, we relate our empirical findings to a more formal theoretical framework by extending
the dynamic search model of Gavazza (2011) to incorporate bankrupt firms that face either (a)
4The correlation between our measures of market thickness and access to finance is 0.10, suggesting that each ofthese channels captures different search frictions that are responsible for the gap between liquidation and reallocation.
5It is particularly important to note that this paper only considers the efficiency of the asset allocation in eachbankruptcy regime. There are, of course, many other aspects of bankruptcy, such as legal fees, creditor recoveries,and worker outcomes, that enter into the overall costs and benefits of liquidation and reorganization which we do notconsider here.
4
agency costs in reorganization, which may lead to inefficient continuation, or (b) forced sales in
liquidation, which could be inefficient due to search frictions. Consistent with our empirical findings,
the model predicts that liquidation leads to lower utilization in thin markets, relative to reorganized
firms. However, this prediction only holds if agency costs in reorganization are low. When agency
costs are high, relatively unproductive firms continue to hold real estate in reorganization and
therefore one would expect that that liquidation will lead to higher utilization by allowing assets
to reallocate to better users. Our empirical results do not match this prediction, as even in thick
markets we do not find that liquidation results in higher utilization than reorganization. Therefore,
our findings are consistent with the idea that when search frictions are high they may hinder
asset allocation in liquidation, and inconsistent with the explanation that high agency costs in
reorganization can explain the results.
This paper contributes to several strands of literature. It is most directly related to Maksimovic
and Phillips (1998), who explore how industry conditions affect the reorganization of large manu-
facturing firms in Chapter 11. More broadly, this paper highlights the importance of local market
characteristics in affecting the consequences of liquidation and reorganization on asset allocation,
and thus contributes to an extensive body of theoretical and empirical literature that discusses the
optimal design and frictions of the bankruptcy process.6 Second, a large literature explores the
existence and implications of fire sales.7 This paper adds to this literature by relying on random
variation that forces liquidation, which allows exploring subsequent reallocation and utilization of
assets separately from reasons that initially lead to the forced sale. Finally, this paper also con-
tributes to the literature that highlights the importance of labor and asset allocation for economic
activity, by studying frictions that may impede reallocation.8
The remainder of the paper is organized as follows. Section I discusses the bankruptcy process.
Section II provides a theoretical framework for our empircal results. Section III discusses the data
construction. Section IV introduces the measurement of asset reallocation and Section V presents
the empirical strategy. Section VI provides the main results in the paper. Section VII discusses the
efficiency implications of our findings and Section VIII concludes.
6Some theoretical examples include Baird (1986, 1993); Gertner and Scharfstein (1991); Aghion et al. (1992);Shleifer and Vishny (1992); Hart (2000), and empirical studies include Hotchkiss (1995); Strömberg (2000); Davydenkoand Franks (2008); Eckbo and Thorburn (2008); Benmelech and Bergman (2011); Chang and Schoar (2013) amongothers.
7For example, see Pulvino (1998, 1999); Ramey and Shapiro (2001); Campbell et al. (2011). Shleifer and Vishny(2011) surveys this literature.
8See, for example, Davis and Haltiwanger (1992); Eisfeldt and Rampini (2006); Hsieh and Klenow (2009); Ot-tonello (2014).
5
I. The Bankruptcy Process
Bankruptcy procedures can be broadly classified into two main categories: liquidation through a
cash auction, and reorganization through a structured bargaining process (Hart (2000)). The U.S.
Bankruptcy code contains both procedures, with liquidation falling under Chapter 7 and reorga-
nization taking place in Chapter 11 of the code. Bankruptcy formally begins with the filing of a
petition for protection under one of the two chapters. In nearly all cases, it is the debtor that files
the petition and chooses the chapter of bankruptcy, although under certain circumstances creditors
can also file for an involuntary bankruptcy. Firms can file for bankruptcy where they are incorpo-
rated, where they are headquartered, or where they do the bulk of their business (see 28 USC §
1408), thereby giving the largest, nationwide firms some leeway in the choice of bankruptcy venue.
However, once a firm files for bankruptcy, it is randomly assigned to one of the bankruptcy judges
in the divisional office in which it files. This random assignment is a key part of our identification
strategy, which we outline below.
Firms that file for Chapter 7 bankruptcy expect to liquidate all assets of the firm, and hence
face a relatively straightforward process, although it can be lengthy (Bris et al. (2006)). A trustee
is put in place to oversee the liquidation of the assets of the firm, and proceeds from the asset sales
are used to pay back creditors according to their security and priority. According to U.S. Court
filing statistics, liquidations are frequent, as about 65% of all business bankruptcy filings in the U.S.
are Chapter 7 filings.
A significant portion of firms that originally file for Chapter 11 bankruptcy also end up in
Chapter 7 through case conversion. Conversion to Chapter 7 occurs when the bankruptcy judge
approves a petition to convert the case. Conversion petitions are typically filed either by a creditor
or the court itself (e.g. by a trustee), accompanied with a brief which outlines why liquidation
will provide the highest recovery for the creditors. As we discuss in Section V, the judge plays an
important role in the decision to convert the case to Chapter 7.9 However, once a case has been
converted, the responsibility to liquidate the estate is passed to a trustee, and thus the judge plays
little role in the reallocation of assets for these cases from that point forward. Meanwhile, firms
that remain in Chapter 11 proceed with the reorganization through a structured bargaining process
governed by specific rights and voting rules defined by the law.10
9We examined court documents for a random sample of 200 cases and found that, on average, a motion to converta case occurs 4 months after the bankruptcy filing. Importantly, in nearly all cases this is the first major motion onwhich a judge rules.
10Specifically, the debtor firm creates a plan of reorganization which outlines which assets will be retained or sold,how the firm will be restructured, and what recoveries creditors will receive. This plan is then distributed to creditorswho vote on the plan. The plan is approved if 2/3 of creditors accept the plan. Because plans are typically negotiatedwith creditors prior to the vote, plan rejections are rare.
6
Importantly, Chapter 11 allows for some or all of the assets of the firm to be liquidated should
that be the outcome of the bargaining process. The key difference from Chapter 7 is that it is not
forced. Assets that are owned by the firm can be sold via “Section 363 sales,” in which some or all
of the firm’s assets are auctioned off while the firm remains in bankruptcy.11 Similarly, in Chapter
11 there is negotiation that determines whether assets that are leased (as much commercial real
estate is) should be retained or returned to their owners. Firms in Chapter 11 have the ability to
choose which leases to accept and which to reject, thereby terminating the contract. In Chapter 7,
leases are automatically rejected, thereby forcing the lessor to find a new tenant. Thus, regardless
of whether an asset is owned or leased, Chapter 11 allows for negotiation surrounding which assets
are kept in the firm, while a new buyer or user must be found for assets in Chapter 7.12
In this paper, we compare asset allocation and utilization across these two bankruptcy proce-
dures. The key difference between the procedures for our purposes is that in Chapter 7 liquidation
all assets are potentially reallocated, while in Chapter 11 reorganization there is negotiation over
which assets remain with the bankrupt firm, or whether that firm survives at all.
II. Theoretical Framework
In this section we provide a theoretical framework to analyze asset allocation between firms. To do
so, we briefly outline an extension to the search model of Gavazza (2011) that models the dynamics
of buyer and seller firms that face search frictions and explores the role of market thickness. Our
extension incorporates two types of bankrupt sellers: (a) reorganized sellers that face agency costs
or (b) liquidated sellers that are forced to sell after the firm shuts down. In section II.A below we
briefly describe the basic setting of Gavazza (2011), and in section II.B we describe our extension and
introduce bankrupt firms. This framework guides the empirical analysis, and aids in understanding
the efficiency implications of our findings, as discussed in Section VII. In this section we provide
the intuition of the model, leaving the formal derivations to Appendix B.
A. Basic Setup
The setup is built on a decentralized market with two-sided search by buyer and seller firms
(Mortensen and Wright (2002); Gavazza (2011)). Firm production relies on a single asset, and
firms are differentiated by heterogeneous productivity that evolves exogenously over time. If firms
own an asset, they choose its level of utilization to maximize the per-period profits from its use, with
11Alternatively, some or all of the assets of the firm can be liquidated through a formal plan of reorganization.Creditors are allowed to vote on these plans.
12A full discussion of the treatment of leases in Chapter 11 can be found in Ayotte (2015).
7
more productive firms choosing a higher level of utilization.13 Firms endogenously select whether
to enter the market and become active buyers or sellers based on their productivity. When a firm
enters the market (either to buy or sell), it pays a search cost to contact other firms willing to trade,
and buyers and sellers meet according to a Poisson process. Once two firms meet and are willing to
trade, they negotiate a trading price.
As illustrated by Gavazza (2011), there exist unique productivity cutoffs for buyers and sellers
that determine their entry into the market. A firm that does not currently own an asset will enter
the market if its productivity increases above the buyers’ cutoff. Similarly, a firm that currently
owns an asset will choose to sell it if its productivity falls below the sellers’ cutoff. The buyers’
threshold is higher than the sellers’ cutoff, and the larger the wedge between these two cutoff values
the slower the reallocation of assets towards more productive users.
Market thickness, defined as the number of potential buyers and sellers in the market, is an
important driver of the wedge between the two cutoffs. The key economic force is that, in a thicker
market, the contact rate between buyers and sellers is higher, assets that are on the market trade
faster, and therefore search costs are lower. Therefore, in thin markets firms choose to hold onto
assets for longer periods, in case their productivity rises in the future, since it is more difficult and
costly to find a buyer. Similarly, potential buyers are less likely to actively search for a seller. As
a result, only very productive firms choose to become buyers and very non-productive firms choose
to become sellers in thin markets, so that the wedge between cutoff values is larger. As the asset
market becomes thicker, the sellers’ cutoff value increases and the buyers’ cutoff value decreases.
Panel (a) of Figure 1 provides a numerical illustration of the equilibrium behavior of buyers and
sellers, showing how the buyers and sellers cutoffs change as market thickness increases.14 Sellers
with a productivity realization below the black dashed line are willing to sell the assets to interested
buyers with productivity levels above the black solid line. Panel (b) of Figure 1 illustrates that
the average holding time - the length of time the current owner holds the asset - declines with
market thickness as expected (see black dotted line), since the probability of a buyer-seller meeting
is higher. Since assets reallocate faster to more productive users in thicker markets, the average
productivity of asset users is higher and so asset utilization is also higher, as shown by the black
dotted lines in Panels (c) and (d).
While the model focuses on market thickness, the same intuition applies in the presence of
financing constraints as in Shleifer and Vishny (1992). When potential buyers are financially con-
13More productive firms will choose a higher level of utilization because they can produce more at lower cost,thereby increasing their optimal utilization level. As pointed out by Gavazza (2011), this is true as long as productivityand utilization are complements in the production function.
14This numerical illustration follows Gavazza (2011), see the Appendix for more details.
8
strained they are less likely to become active buyers, sellers face essentially thinner asset markets
and higher search costs and are therefore inclined to hold on to the assets for longer periods.
B. Introducing Bankrupt Sellers
We introduce two types of bankrupt sellers in the model: liquidated sellers and reorganized sellers.
We assume for simplicity that the number of bankrupt sellers is small and thus they do not affect
the equilibrium behavior of regular buyers and sellers in the economy. Their objective functions,
however, differ from regular sellers in important ways.
In liquidation, the firm ceases to exist, and sellers can no longer use the asset to generate
intermediate profits. This makes liquidated sellers particularly eager to sell the asset relative to
regular sellers. Since liquidated sellers are likely to become active sellers in almost any level of
market thickness, they are particularly exposed to market conditions. When markets are thin,
search times are lengthy, during which liquidated sellers do not utilize the asset.
Meanwhile, in reorganization the firm continues to operate and can still generate intermediate
profits from the asset, thereby avoiding impatient sales and unused assets. However, because of
agency problems, managers may continue to hold on to the assets, despite the availability of more
productive potential buyers, because the cost of doing so is mostly borne by senior claimants (Baird
(1986); Jensen and Meckling (1976)).15 We model agency costs as private benefits that managers
receive from continuing the firm and holding on to the asset. The implications of this assumption
can be illustrated in Panel (a) of Figure 1. Reorganized sellers’ cutoff for becoming active sellers
(red dashed line) is lower than regular sellers, and therefore they are willing to become active sellers
only if they experience a particularly bad productivity shock. Therefore, they hold on to the assets
inefficiently, relative to regular sellers, and may pass up opportunities to reallocate the asset to more
productive users.16
The differences between liquidated and reorganized sellers can be illustrated through the average
asset holding time. As shown in Panel (b) of Figure 1, the asset holding time of liquidated sellers
is much shorter than regular sellers, while reorganized sellers hold the asset longer than regular
15He and Matvos (2015), for example, provide a model where debt exists to limit excessive continuation by causingfinancial distress. However, many question if bankruptcy actually limits inefficient continuation. Chapter 11 allowscurrent managers to retain control of the firm and gives them the exclusive right to propose a plan of reorganization.In addition, voting rules can allow junior claimants or management to push for continuation even when their claims areout of the money, which they might do to preserve their jobs or to gain some recovery value (Bradley and Rosenzweig(1992), Hart (2000)). Thus, financial distress creates a larger wedge between ownership and control, while bankruptcylaws serve to actually strengthen the control of current managers and equity holders in the reorganization regime.Other papers that provide theoretical arguments for excessive continuation based on agency costs include Gertnerand Scharfstein (1991) and Bebchuk and Chang (1992).
16For liquidated sellers, markets need to be extremely thin to deter them from actively searching for a buyer, andtherefore their threshold is not presented in Panel (a) of Figure 1.
9
sellers. This is because liquidated sellers almost always actively search for a potential buyer and
therefore sell the asset much more quickly. In contrast, reorganized sellers, due to the agency cost,
are less likely to actively seek a buyer, and therefore hold the asset longer. Interestingly, the gap
in holding time between liquidated and reorganized sellers narrows as the market becomes thicker,
as reorganized sellers become more willing to sell when search frictions decline. This motivates our
first hypothesis:
Hypothesis #1: Asset holding time is shorter in liquidation than in reorganization. However,
the gap declines as market thickness increases.
We use the model to calculate asset utilization over the lifetime of an asset. For liquidated
sellers, the assets are not used until they are sold, while reorganized sellers can utilize the asset
until they decide to sell. The long-run utilization, therefore, depends on the holding time, and
also on the productivity of the buyer after reallocation takes place. Panel (c) plots the average
asset utilization for regular sellers (black dotted line), liquidated sellers (blue line) and reorganized
sellers (red dashed line). In thin markets, liquidation leads to lower utilization than reorganization.
However, as markets become thicker, utilization in liquidation grows faster than in reorganization,
such that there is a crossing point at which asset utilization in liquidation becomes greater than
in reorganization. This is because liquidated sellers can reallocate the asset more quickly to more
productive buyers when markets are thick.17 In contrast, reorganized sellers are slower to sell and
more insulated from the markets due to the private benefits from holding the asset and the ability
to generate intermediate profits.
The model gives similar predictions with respect to the average life-long productivity of the
asset users, as illustrated in Panel (d). In thin markets, the average productivity of users is higher
in reorganization because it is difficult for liquidated sellers to find a buyer. However, as markets
become thicker, the average productivity of assets in liquidation increases sharply and the gap
between reorganization and liquidation productivity narrows until it becomes higher in liquidation.18
This leads to a second testable hypothesis:
Hypothesis #2: Asset utilization and productivity increase with market thickness for both
17As markets get thicker, the holding time of liquidated sellers decline because more potential buyers are becomingactive. This is hard to see in Panel (b) of Figure 1 since the holding time of reorganized sellers is much larger andwe rely on the same scale. However, in the numerical illustration, the holding time of liquidated sellers drops from1.3 in thin markets to 0.3 in thick markets. That is, a four fold decline in the holding time of the asset.
18It is important to note that these predictions are based on a model with only low agency costs. If agencycosts are high, then utilization and productivity of reorganized assets is always lower than that of liquidated assets,even in thin markets, and the gap increases with market thickness. The empirical findings are inconsistent with thepredictions of high agency costs scenario, as we discuss this in detail in Section VII below.
10
reorganized and liquidated sellers. However, since liquidated sellers are more sensitive to changes
in market thickness, the increase is faster in liquidation than in reorganization.
This hypothesis implies that if utilization and producitivity are higher in reorganization in thin
markets, relative to liquidation, this gap should decline or dissapear as markets become thicker.
As mentioned above, to ease the exposition we have described the implications of the model
with regards to market thickness. However, similar results hold if one varies the accessibility of
capital, as in Shleifer and Vishny (1992), since financing constraints will limit the set of potential
buyers, and sellers will face thinner markets with higher search costs. Accordingly, in Section VI.B
we test empirically how the gap between liquidation and reorganization changes both with respect
to market thickness and access to capital.
Importantly, this theoretical framework also helps to interpret the relative importance of agency
costs and search frictions that face bankrupt sellers. Specifically, when agency costs are high relative
to search costs, the utilization and productivity of reorganized sellers can be reduced to levels below
those of liquidated sellers since reorganized sellers will resist reallocating the assets. Thus, comparing
the utilization of assets in liquidation and reorganization across levels of market thickness can help
us understand which frictions affect asset allocation in bankruptcy. We discuss this issue in detail
in Section VII below.
III. Data
A. Bankruptcy Filings
We gather data on Chapter 11 bankruptcy filings from LexisNexis Law, which obtains filing data
from the U.S. Courts system. This data contains legal information about each filing, including the
date the case was filed, the court in which it was filed, the judge assigned to the case, an indicator of
whether the filing was involuntary or not, and status updates on the case. From the status updates,
we are able to identify cases that were converted to Chapter 7. The LexisNexis dataset contains a
few bankruptcies beginning as early as 1980, but coverage is not complete in these early years as
courts were still transitioning to an electronic records system. We begin our sample in 1992, when
LexisNexis’ coverage jumped to over 2,000 bankruptcy filings per year (from 450 in 1991) across 70
different bankruptcy districts (out of 91). By 1995, LexisNexis covers essentially 100% of all court
cases across all bankruptcy districts.19 The comprehensive nature of the LexisNexis data makes this
one of the largest empirical studies on bankruptcy to date, including both public and private firms
19Iverson (2015) provides more details of the LexisNexis data.
11
from all bankruptcy districts and across all industries. We end our sample with cases that were filed
in 2005 so as to be able to track bankrupt firms for a five-year period after the bankruptcy filing.
B. Census Data and Measures of Local Market Characteristics
We match bankruptcy filings from LexisNexis to their establishments in the U.S. Census Bureau’s
Business Register (BR), which we then link to the Longitudinal Business Database (LBD). The LBD
includes all non-farm tax-paying establishments in the U.S that employ at least a single worker. In
the LBD, an establishment is a physical location where economic activity occurs. This serves as the
main unit of observation in our study.
We match the bankruptcy filings from LexisNexis to the BR using the employer identification
number (EIN), which is contained in both datasets. Importantly, each legal entity of a firm can have
a separate EIN, and thus there can be multiple EINs (and multiple bankruptcy filings) for each firm.
Further, an EIN can have multiple establishments connected to it in the LBD. We match bankrupt
EINs to all establishments in the BR in the year of the bankruptcy filing to form our initial sample of
bankrupt plants. This sample is then reduced due to missing addresses (which are necessary to track
economic activity at a location), resulting in a final sample of 129,000 establishments belonging to
28,000 unique firms.20
Table 1 presents summary statistics for our final sample. Panel A shows that the average firm in
our sample has 4.7 establishments and employs 169 individuals. In total, firms employ 4.7 million
individuals at the time of the bankruptcy filing. Approximately 40% of the bankruptcy filings in our
sample convert to Chapter 7 liquidation. Further, there are stark differences between firms that stay
in Chapter 11 and those that are converted to Chapter 7. The average Chapter 11 firm has nearly
three times as many establishments and over four times as many employees.21 These differences are
apparent also at the level of the plant, where plants of Chapter 11 firms employ almost 50% more
workers than those of firms that convert to Chapter 7. In addition, Chapter 11 firms have higher
payroll per employee ($26,000 per year versus $20,200 at Chapter 7 firms) and are about two years
older than Chapter 7 firms. The differences between Chapter 11 and Chapter 7 firms highlight the
importance of selection into bankruptcy regimes, and hence the need for identification in assessing
the impact of the regimes.
Panel B of Table 1 shows the industry distribution of plants and firms in our sample. Firms are
distributed across all industries, with wholesale and retail trade and services making up the largest
20We provide extensive details of the matching process and sample selection in Appendix C.21Census disclosure rules prohibit reporting medians or percentiles, but Table 4 reports the number of firms in
different size categories to give a better representation of the size distribution of our sample.
12
portions of the sample.22 We do not see large differences in the share of firms that are liquidated
across different sectors; roughly 40% of firms are converted to Chapter 7 in all industries.
In Section VI.B, we explore two measures of heterogeneity of local market characteristics that
are related to search frictions: market thickness and access to capital. Following Gavazza (2011),
we first focus on market thickness as a principal driver of the ability to redeploy assets. Given that
reallocation is typically done locally and within the same industry (as we show below), we expect
that counties which contain many firms in the same or similar industries as the bankrupt plant will
have lower search costs and hence a higher probability of finding a user of the vacated real estate.
We use the full LBD to measure market thickness for industry i in county c in year t as
Thicknessict =X
j
⌧ijsjct,
where ⌧ij is the observed probability across our full sample that a plant in industry i transitions
to industry j after closure, and sjct is industry j’s share of total employment in county c in year
t.23 Thicknessict is essentially a weighted index of market concentration, where each industry is
weighted by ⌧ij . ⌧ii, the probability that a plant remains in the same industry, is substantially
higher than any other ⌧ij for all industries, implying that it is often difficult to transition an asset
to a new industry. Indeed, by weighting by ⌧ij , this measure of thickness accounts at least in part
for two types of asset specificity. First, for some industries the physical characteristics of the asset
are not suited for other uses, such as in the case of a factory, making reallocation difficult. Second,
it can be that specific locations are better suited for particular businesses due to their proximity
to suppliers or customers, as in the case of tech firms in Silicon Valley or retail stores in shopping
malls. Thicknessict is highest when a given county has a high concentration of plants in the same
or related industries, where there are many potential buyers who can use the asset for its intended
purpose without having to overcome either form of asset specificity. Indeed, the same county can
have both a high thickness measure for one type of asset and a low thickness measure for another,
depending on the local industrial composition. Interestingly, in Appendix Table A.9 we show that
market thickness is quite similar across industries, suggesting that the variation in this measure
stems from geographic variation within industry, rather than across industries. In Panel A of Table
1, we also show that levels of market thickness are similar for both reorganized and liquidated firms.
Second, we focus on access to finance as a determinant of asset reallocation, as in Shleifer
and Vishny (1992). Because the majority of new occupants of bankrupt assets are local or new
22The number of firms in Panel B sums to more than 28,000 because some firms have plants in different industries.23Results remain unchanged if we define sjct as the share of plants in industry j rather than the share of employ-
ment.
13
firms (as we discuss below), we expect that small business loans will be the principal source of
capital for these firms (Petersen and Rajan (1994)). Accordingly, we use the share of loans going
to small businesses in a county as a proxy for access to finance. We measure this share using
the Community Reinvestment Act (CRA) disclosure data from the Federal Financial Institutions
Examination Council (FFIEC), which contains data on loan originations by commercial banks for
loans under $1 million.24 Specifically, we proxy for access to capital by measuring the share of small
business loan originations going to small businesses, defined as firms with less than $1 million in
annual gross revenue.25 In Panel A we find that the share of small business loans in regions of
reorganized firms is similar to those in regions of firms that were converted to liquidation.
IV. Asset Allocation Measurement
A. Tracking Real Estate Assets Over Time
In this section we describe the construction of geographical linkages that track bankrupt firms’
real estate locations over time. We track assets even when plants are sold or shut down, thereby
capturing whether real estate is occupied (by either a bankrupt firm or a different occupier), and if
so, how intensively it is utilized, as captured by the asset’s total employment. To do so, we rely on
the Census LBD, which covers all non-farm, private sector establishments in the United States. A
significant benefit of the LBD is that it captures the location of tax-paying establishments, thereby
reporting the users of real estate assets. This allows us to carefully explore asset reallocation through
the evolution of asset occupiers, and asset usage, regardless of whether the property is owned or
leased.26
To track real estate occupancy and employment outcomes over time, we create a careful address
matching algorithm to link addresses over time. First, we clean all addresses and address abbrevia-
tions using the United States Postal Service formal algorithm.27 Then, for each shut-down plant, we
24The CRA requires banks above a certain asset threshold to report small business lending each year. Duringour sample period, the asset threshold was $250 million. Greenstone et al. (2014) estimate that CRA eligible banksaccounted for approximately 86% of all loans under $1 million.
25Following Greenstone et al. (2014), we define small business loans as those up to $1 million, and small businessesas firms with less than $1 million in annual gross revenue. Ideally, we would measure the share of all lending thatgoes to small firms, rather than just the share of loans under $1 million, but county-level data on all loans is notavailable. Given that over 50% of loans less than $1 million go to large firms, it is likely that nearly all loans greaterthan $1 million go to large firms, and thus the share of CRA loans going to small businesses is a reasonable proxyfor the share of all lending going to small businesses.
26An alternative approach would be to rely on real estate transactions, following changes in asset ownership.However, such an approach cannot identify whether assets are directed to different uses if reallocation occurs throughleases. Moreover, this approach cannot identify when assets are vacant, and the extent to which the assets are used.
27See the following link (valid as of October 2016) for details of the postal addressing standards used:http://pe.usps.gov/text/pub28/
14
attempt to match its address with subsequent LBD years (up to five years following the bankruptcy
filing), to track the next occupier of the real estate location.28 We also match to previous years
of the LBD up to three years prior to the bankruptcy to verify that the pre-trend in employment
is unrelated to the judge instrument, as discussed in Section V. Our address matching algorithm
forces a perfect match on both zipcode and street numbers for each location, and then allows for
(almost perfect) fuzzy matching on street name and city name. The details of the address matching
algorithm are provided in Appendix D.
With these geographical linkages, we categorize each plant outcome in the following manner.
First, if a plant continues to operate (i.e. has positive payroll) after the bankruptcy filing under its
original ownership we classify the plant as “continued.” Second, if a real estate location is occupied
and active, and is used by a different firm from the original bankrupt occupier, we classify it as
“reallocated.”29 Such reallocation may not necessarily take place immediately. Therefore, in a given
year, we say that a plant is “vacant” if the original plant has previously shut down and no active
plant is currently occupying the real estate location.
We also link addresses of non-bankrupt plants in the same county as the bankrupt plants to
create a benchmark of plant occupancy and utilization. To do so, we create a 5% random sample
of all plants in the LBD in the same county and year as each bankrupt plant and track the asset
allocation and utilization of this set of over 4 million establishments in exactly the same manner as
the bankrupt establishments.30 This not only provides a benchmark against which to compare our
results, but also provides an estimate of local area vacancy rates which can proxy for the opportunity
costs of vacancy, as discussed in detail in Section VII below.31
28The LBD includes plant identifiers that link establishments over time. These plant linkages broadly rely onname and address matching (see Jarmin and Miranda (2002) for a detailed description of the construction of theplant linkages). Hence, plant linkages are maintained as long as a plant remains active under existing ownershipor is sold and the new owner keeps the same plant name and address. Otherwise, the plant identifier link is notmaintained. Our goal is to construct location-based linkages which are robust to any change in name, and followplant locations more broadly. Importantly, in our sample the standard LBD linkages account for only about 25% ofreallocation, while the geographical linkages we construct account for the remaining 75%.
29A real estate location could be sold back to its original owner under a new legal entity, in which case it would bemarked as “reallocated” in our data even though it has not truly been reallocated, as we lack the ability to determineif the new legal entity is related to the original owner. Strömberg (2000) shows that these “sale-backs” are relativelycommon in Sweden. However, in examining 100 randomly-selected bankruptcy cases in our sample we did not findany instances of asset sale-backs in the U.S.
30We also construct local benchmarks based on the same 3-digit NAICS and county as the bankrupt plant. Inunreported results we find essentially identical figures.
31Because address matching is inherently imperfect, we conduct a variety of checks, including manually examiningmatches and comparing to external data, to ensure that our match quality is high and that our results are notdependent on matching issues. These checks are discussed in Appendix Section D.D. Further, Appendix Table A.8shows that our results are robust to excluding plants for which matching is less precise, such as office buildings orshopping malls with many establishments.
15
B. Stylized Facts about Asset Allocation in Bankruptcy
In this paper we construct measures of asset allocation and utilization of real estate assets. Given
the novelty of the measures, in this section we describe three stylized facts that guide our main
analysis in Section VI below.
Stylized Fact 1: Asset Reallocation is Prevalent in Both Bankruptcy Regimes
In Panel A of Figure 3, we explore whether plants continue to be operated by their initial users
following the bankruptcy filing under either liquidation or reorganization. We find that when a
bankruptcy filing is converted to Chapter 7, only 54% of plants continue to operate under original
ownership after one year, and only 8% by year three. While it is expected that liquidated plants
will not continue, non-continuation is also prevalent in reorganization. Specifically, 70% of Chapter
11 plants continue after one year, and that figure drops to 39% by year three and 26% by year five.
In comparison, non-bankrupt benchmark plants that are located in the same county have a survival
rate of 71% after three years, and 59% by year five.
Panel B of Figure 3 provides evidence on the importance of reallocation in bankruptcy. The figure
shows the probability that a location is occupied by any firm across the two bankruptcy regimes.
A comparison of this panel with the figures in Panel A illustrates the extent to which assets are
reallocated. Five years after bankruptcy, only 26% of reorganized plants continue with the original
bankrupt firm, but 69% are occupied, meaning that 43% of these locations are reallocated to new
users. Meanwhile, occupancy rates among liquidated plants are 55% by year five, and the occupancy
of these plants is entirely due to reallocation.32 However, the occupancy rates of both bankruptcy
regimes remain below that of the non-bankrupt benchmark for all years. A similar picture arises
when exploring utilization in terms of total employment, as illustrated in Panel C of Figure 3. After
accounting for reallocation to new users, employment at reorganized firms drops to about 70% of its
pre-bankruptcy level by year three and remains close to that level. Meanwhile, employment drops
quickly at liquidated plants and then recovers to just over 60% of pre-bankruptcy employment by
year five after bankruptcy. These are strikingly different to employment at benchmark plants, whose
employment grows slightly over time.33 Of the 3.25 million workers employed at bankrupt locations
32Even after accounting for reallocation, vacancy rates are still over 30% in year 5 among bankrupt firms, and 20%among the local benchmark of non-bankrupt firms. For reference, statistics collected by the National Association ofRealtors indicate that commercial real estate vacancy rates nationwide average over 10%, with levels as high as 20%not being uncommon (see http://www.realtor.org/reports/commercial-real-estate-outlook, link valid as of January2016). Bankrupt firms are more likely to reside in poorly performing regions, and assets may be more likely to beneglected, thus explaining the higher vacancy rates. In a series of papers, Steven Grenadier (Grenadier (1995, 1996))finds evidence for vacancy rates as high as 30% in the Denver and Houston areas in the 1980s, and shows that thelevel of equilibrium vacancy rates is predominately determined by local factors. Moreover, he illustrates a significantpersistence in vacancy rates in commercial real estate.
33The occupancy rate of benchmark plants naturally falls over time since by definition all benchmark plants areoccupied in year 0. However, overall employment increases at benchmark plants as non-vacant establishments grow
16
by year 5, more than 2 million workers are employed at locations that have been reallocated to
new users. Thus, by occupancy or employment, asset reallocation plays an important role in the
utilization of these bankrupt assets.34
Stylized Fact 2: Search Costs and Asset Specificity Matter for Reallocation
We find that search costs and asset specificity are important features of the reallocation process
in bankruptcy. In Panel C of Table 1 we explore the characteristics of reallocated bankrupt plants.
We find that most assets are reallocated to local firms, either newly created businesses (52.0%)
or existing firms that already have at least a single plant in the same county (34.4%). Non-local
entrants account for only a small fraction (13.6%) of total reallocations. This is especially true for
liquidated plants, where new entrants account for 70.4% of all reallocation, and non-local entrants
only make up 7.4%. We also find a high degree of reallocation within industries, as the probability
that reallocated asset will remain within the same 3-digit industry NAICS is 46.4%. Note that if
assets were to randomly transition between industries, the probability of reallocating within the
same 3-digit industry would be about 2.2%, given the size distribution of industries in the LBD.
Interestingly, Panel C also shows that liquidated plants are about 11 percentage points less likely
to remain in the same 3-digit NAICS than reorganized plants. These results are consistent with
the literature documenting the importance of asset specificity and search cost in asset reallocation
(Ramey and Shapiro (2001); Eisfeldt and Rampini (2006); Gavazza (2011)), as discussed above.
Stylized Fact 3: Industry and Local Economic Conditions Affect Reallocation
Finally, we find that industry and local economic conditions are important in determining the
degree of asset reallocation. Table 2 reports regression results in which we limit the sample to
plants that do not continue with the bankrupt firm, and explore what affects the probability that
real estate assets will be reallocated and utilized by a new owner as opposed to remaining vacant.
The dependent variable is an indicator equal to one if a new establishment occupies the real estate
location within five years of the bankruptcy filing, and zero if the plant was closed but not replaced.
In column 1, we find that county-level characteristics are significant predictors of asset reallo-
cation. In particular, we find that being located in a county with a high total number of plants,
high economic growth (measured by three-year employment growth in a county), and high payroll
per employee, are significantly correlated with a higher probability that a discontinued plant will
be reallocated.
We find that industry-level conditions matter as well in column 2, which illustrates that real
by more than the drop in employment at vacant locations.34Relatedly, we find that reallocation, when it takes place, occurs almost immediately in both bankruptcy regimes.
Conditional on transitioning to a new user, approximately 65% of plants are reallocated in the same year they areshut down. The probability that the real estate is redeployed falls drastically in subsequent years.
17
estate in high-growth industries is more likely to be reallocated. In column 3, we also report industry
dummies to illustrate heterogeneity across industries in reallocation likelihood. For example, real
estate in accommodation, food and entertainment, is much more likely to be reallocated (conditional
on plant closure) relative to the mining and construction omitted category. This evidence suggests
that the degree of asset specificity, and the number of potential buyers for commercial real estate
may vary across industries.
In columns 4, 5, and 6 of Table 2, we control simultaneously for county-level and industry char-
acteristics, and we vary the set of fixed effects that we include in the estimation. All county-level
characteristics remain highly significant in these regressions as well as industry fixed effects, but
the effect of industry growth rates falls to zero. Overall, the results highlight the cross-industry
variations in reallocation propensities, and the importance of local economic conditions. This mo-
tivates our focus on local market conditions, and in particular the presence of local firms in similar
industries, as important determinants of reallocation in bankruptcy.35
V. Identification Strategy
A. Empirical Design
Identifying the effect of Chapter 7 liquidation on asset allocation relative to Chapter 11 reorgani-
zation is challenging given the inherent selection into bankruptcy regimes. Firms filing directly for
Chapter 7 may have worse prospects, and this will be reflected in the way their assets are allocated
and subsequently utilized. To mitigate the selection, we focus only on firms that filed for Chapter
11 reorganization, and exploit the fact that a significant fraction (40%) of these firms are converted
to Chapter 7 liquidation subsequently. Hence, the baseline specification of interest is:
Yp,i,t+k = ↵+ � · Liquidationp,i,t + �Xp,i,t + ✏p,i,t+k
where p indexes an individual plant real estate used by firm i, t is the year of the bankruptcy filing,
and k indexes the number of years after bankruptcy (ranging from one to five). The dependent
variable Yp,i,t+k is a measure of post-bankruptcy plant outcomes and real estate asset utilization
such as the total number of workers employed at real estate p in year t + k. We are interested in
estimating �, which captures the impact of conversion to liquidation on Yp,i,t+k, after controlling
for a set of firm- and plant-level variables, Xpit, such as pre-bankruptcy filing employment and
plant age. We index Liquidationp,i,t as occurring in year t, since the decision of whether the case
35The fact that reallocation is strongly related to local market and industry conditions also supports the validityof our address-matching procedure. If the matching were noisy, such strong patterns would not emerge in the data.
18
is converted to Chapter 7 liquidation or remains in Chapter 11 reorganization is typically taken in
the bankruptcy filing year.36 Under the null hypothesis that liquidation has a similar effect on asset
utilization as reorganization, � should not be statistically different from zero.
Even within Chapter 11 filers there may be a significant amount of selection among firms that
convert to Chapter 7 liquidation. Table 1 illustrates this point, as firms converted into Chapter 7
liquidation tend to have a smaller number of plants, employ fewer workers, and are slightly younger.
Therefore, to identify the causal effect of liquidation on plant outcomes and asset allocation, we
rely on judge heterogeneity in their propensity to convert Chapter 11 filings to Chapter 7 as an
instrumental variable.37 This instrument does not rely on differences in actual bankruptcy laws,
as the bankruptcy code is uniform at the federal level. Rather, the instrument makes use of the
fact that bankruptcy judges’ interpretation of the law varies significantly (LoPucki and Whitford
(1993); Bris et al. (2006); Chang and Schoar (2013)).
Bankruptcy judges work in 276 divisional offices across the United States, each of which pertains
to one of 94 US Bankruptcy Districts. A firm filing for bankruptcy may choose to file either where
it is (1) headquartered, (2) incorporated or (3) does most of its business, thereby giving the largest
firms some leeway in the bankruptcy venue. However, once a filing is made in a particular division,
judge assignment is random.38 We can then rely on this random assignment to generate exogenous
variation in the probability that a given case is converted, since judges vary in their propensity to
convert filings. To implement the instrumental variables approach, we estimate the following first
stage regression:
Liquidationp,i,t = ⇢+ ⇡ · �j + �Xp,i,t + �d,t + µk + ✏p,i,t
where Liquidationp,i,t is an indicator variable equal to one if the bankruptcy case was converted
to Chapter 7 liquidation and zero otherwise. Importantly, we include division by year fixed effects,
�d,t, to ensure that we exploit judge random variation within a division-year. We also include plant-
level controls Xp,i,t and industry fixed effects, µk. The coefficient on the instrumental variable, ⇡,
represents the impact of judge j’s tendency to convert a case to Chapter 7, �j , on the probability
that a case is converted to Chapter 7 liquidation. We estimate �j as the share of Chapter 11 cases
36To verify this, we examined the court documents of 200 randomly selected cases in our sample, and found thatfor the median case the time between case filing and a decision on whether the case will remain in Chapter 11 or beconverted to Chapter 7 is 4 months.
37This approach was pioneered by Kling (2006), and has been applied in a variety of settings (Doyle Jr (2007); DoyleJr. (2008); Maestas et al. (2013b); Di Tella and Schargrodsky (2013); Dahl et al. (2014); Galasso and Schankerman(2015); Chang and Schoar (2013); Dobbie and Song (2015)).
38As an example, consider the bankruptcy district of New Jersey, which is divided into 3 divisions: Camden,Newark, and Trenton. The Local Rules of the New Jersey Bankruptcy Court lay out exactly which counties pertainto each division, and firms must file in the division “in which the debtor has its principal place of business.” Once acase is filed in a particular division, the Local Rules state that “case assignments shall be made by the random drawmethod used by the Court.”
19
that judge j ever converted to Chapter 7, excluding the current case. This standard leave-one-out
measure deals with the mechanical relationship that would otherwise exist between the instrument
and the conversion decision for a given case and follows the previous literature that uses the random
assignment of judges as an instrument (e.g., Doyle Jr (2007), Maestas et al. (2013a), Galasso and
Schankerman (2015) among others).39 The second stage equation estimates the effect of liquidation
on plant outcomes:
Yp,i,t+k = ↵+ � · dLiquidationp,i,t + �Xp,i,t + �d,t + µk + ✏p,i,t+k
where dLiquidationp,i,t are the predicted values from the first stage regression. In all regressions we
cluster standard errors at the division-by-year level, to account for any correlation within bankruptcy
court.
If the conditions for a valid instrumental variable are met, � captures the causal effect of Chapter
7 liquidation on plant outcomes and asset allocation, relative to reorganization. It is important to
note that the estimates in the instrumental variables analysis are coming only from the sensitive
firms - those firms which switch bankruptcy regimes because they were randomly assigned a judge
that commonly converts cases (Imbens and Angrist (1994)). Clearly, there are some firms that will
stay in Chapter 11 no matter the judge and there are other firms that will convert to Chapter 7
regardless of the judge. Thus, the instrumental variables estimates only capture the local average
treatment effect on the sensitive firms, and should be interpreted as such. We discuss the set of
sensitive firms in our sample below.
B. Judge Heterogeneity and Conversion to Liquidation
For the instrument to be valid, it must strongly affect the likelihood of conversion to Chapter 7
liquidation. This can be illustrated in Figure 4, which plots the nonparametric kernel regression
between the probability that a case is converted to liquidation and �j , the share of Chapter 11 cases
that a judge ever converted, after purging both variables for division-year fixed effects, industry fixed
effects, and all control variables in Xp,i,t. We confirm this evidence in our first stage regression,
presented in Table 3, which demonstrates that there is a strong and tightly estimated relationship
between the instrument and the probability of conversion to liquidation, even after introducing a
comprehensive set of controls.
39In Table A.2 in the Appendix we show that the results are unchanged if we define the instrument as the share ofcases that judge j converted to Chapter 7 in the five years prior to the current case. We also show that judge leniencyis quite consistent over time, as judge decisions in the first half of their tenure strongly predict their decisions in thesecond half of their tenure with a coefficient close to one, as shown in panel C of Appendix Table A.2. This furthermotivates the use of the standard leave-one-out measure of judge leniency in the main analysis.
20
In column 1 of Table 3 the unit of observation is a bankruptcy filing. The result illustrates
that the instrument, share of other cases converted, is strongly and significantly correlated with
conversions to liquidation. In particular, a one standard deviation (12.9%) increase in our instrument
increases the likelihood of conversion by 7.49%, a 18.37% increase from the unconditional propensity
of 40.74%.
In the remaining columns of Table 3, and in fact in the entire analysis below, the unit of
observation is at the plant location level rather than the bankruptcy case level. In these regressions
each observation is weighted by the inverse of the number of plants operated by the firm, to ensure
that each firm receives the same weight in the regression and to avoid overweighting large bankruptcy
cases. In column 2 we repeat the specification in column 1, and verify that the first stage results are
identical to column 1 in which the unit of observation is at the bankruptcy case level. In column
3 we add additional control variables, such as the plant age and number of employees per plant at
the year of the bankruptcy filing.40 The results remain unchanged. In Table A.2 of the Appendix
we illustrate that the results are robust to alternative instrumental variable specifications discussed
above. In all specifications, the F-stat is above 100, well above the required threshold of F = 10 to
alleviate concerns about weak instruments (Staiger and Stock (1997)).
Another identifying assumption is monotonicity, which requires that the assignment of a judge
has a monotonic impact on the probability that a given Chapter 11 case is converted into Chapter
7. This means that while the instrument may have no effect on some firms, all those who are
affected are affected in the same way. The assumption would be violated if we observe certain
types of firms for which the likelihood of conversion increases after being assigned to a given judge,
and other firms treated with the same judge for which the likelihood of conversion decreases. This
implies that the first stage estimates should be non-negative for all subsamples. In unreported
regressions, we estimate the first stage regression for samples split by the median for the following
characteristics: number of employees at plant or firm, number of plants in firm, county, or industry,
plant age, three-year employment growth in county or industry, and payroll per employee in county
or industry. We also estimated the first stage separately for multi-state and multi-county firms
and single-state and single-county firms, since firms in multiple counties or states might be able to
“forum shop” for a different bankruptcy venue. As we can see from Table A.3 in the Appendix, the
estimates are positive and sizable in all subsamples, in line with the monotonicity assumption. We
40Surprisingly, ln(employees at plant) is positively related to the likelihood of liquidation, while ln(total employeesat firm) is negatively related. This is because these two covariates are the same for all single-establishment firmsin our sample, making them somewhat multi-collinear. If ln(total employees at firm) is omitted from the regression,the coefficient on ln(employees at plant) becomes negative and significant, as expected. However, the instrument isorthogonal to both of these variables, so it is unaffected by either of these controls, and we include both in order tocontrol for plant-level and firm-level characteristics in the second stage.
21
test the first stage on further subsamples as discussed in the next section, and continue to find that
the coefficient is positive in all sample splits.
C. Characterizing Marginal Firms in the Bankruptcy System
In this section, we characterize the marginal firms in the bankruptcy system by running the first
stage regression on various subsets of the data. This analysis accomplishes two goals. First, it helps
with understanding the scope of the local average treatment effects of our analysis, by exploring
which firms are likely to be sensitive to the instrument. Second, from a policy perspective it shows
which firms would be most affected by a policy change in bankruptcy.
As pointed out by Maestas et al. (2013b), when the treatment variable is binary and the instru-
ment varies between 0 to 1, the size of the population that is marginal is equal to the first stage
coefficient. In our case, moving from the most lenient judge, who converts zero cases, to the most
strict, who converts all cases, would shift 58.1% of firms into liquidation, because the coefficient on
share converted is 0.581, as illustrated in Table 3.41 Building on this insight, by running the first
stage regression on different subsamples we can determine the share of firms in each category that
are “compliers” - that is, firms that are sensitive to the judge assigned to the case. Table 4 gives a
summary of our findings.
We first split by size of the firm, looking at firms with 0-5 employees, 6-25 employees, 26-100
employees, 101-1,000 employees, and 1,000+ employees. Among these groups, the raw share of
firms converted to liquidation, as noted in column 3, is 44%, 44%, 41%, 29%, and 13%, respectively,
showing even among larger firms a substantial number are liquidated. In column 4 we display the
coefficient on share converted from the first stage regression. With the exception of firms over 1,000
employees, the coefficient is large and highly significant for each subgroup, demonstrating that a
high proportion of firms with less than 1,000 employees is sensitive to judge biases. Interestingly,
the share of firms that are marginal is non-monotonic in size. Just under 50% of firms with 0-5
employees are sensitive to judge assignment, and this proportion grows to 88% of firms with 26-100
employees. The proportion then decreases again, such that 52% of firms with 101-1,000 employees
are compliers, and only 26% of firms with over 1,000 employees are sensitive to the judge. It is
perhaps unsurprising that only few of the largest firms are compliers, as only 13% of these firms
are converted to Chapter 7 overall and presumably the stakes are large enough in these cases that
judicial preferences are of less consequence.42 However, we also point out that our sample contains41Of course, moving from the most lenient to the most strict judge is an extreme shift and is only a hypothetical
exercise. Because most judges are near the center of the distribution, randomly reassigning firms to judges wouldresult in far fewer firms moving to other bankruptcy regimes.
42When firms of this size liquidate, it is more common for them to do so within Chapter 11, rather than convertingto Chapter 7.
22
only 1000 firms with over 1,000 employees, and for this reason there is not much statistical power
to identify the effect of the judge after the inclusion of division by year fixed effects.
We can also estimate the share of firms that would always be converted to liquidation regardless
of the judge. We estimate this share of “always takers” by using the first stage regression coefficients
to predict the probability of liquidation if �j = 0, meaning the firm was assigned to the most lenient
judge in the sample. We then average the predicted probability of liquidation under the most lenient
judge to estimate the share of always takers in each subgroup.43 The findings are both intuitive
and interesting. In column 6, we find that the smallest firms have the highest share of always
takers, consistent with the intuition that the smallest firms are the least likely to need Chapter 11
protection. Meanwhile, the fraction of always takers declines substantially for all firms with more
than 25 employees.
Combining this with the analysis of the share of compliers in column 4, we can characterize the
different size categories as follows. Among firms with less than 25 employees, about one quarter
are converted to liquidation regardless of the judge and about half are marginal. Nearly all of the
middle-sized firms are compliers, with only about 6% being always takers and an additional 6%
being “never takers” - firms that would not be converted even under the strictest judge. Meanwhile,
hardly any of the largest firms are always takers and a relatively small portion are compliers, leaving
a large fraction of never takers.
Table 4 also displays first stage coefficients and the percentage of always takers for the sample
splits of market thickness and access to finance. We find little difference in the first stage when we
split by market thickness, with a 55.7% (19%) share of compliers (always takers) in thick markets
compared to 60.3% (16%) in thin markets. On the other hand, we see slightly larger differences
when we split the sample by access to finance, where 66.4% of firms are sensitive to the judge
assignment in markets with high access to finance (14% always takers) compared to 48.4% in low
access to finance areas (23% always takers).
D. The Exclusion Restriction Condition
Our identification strategy is designed to overcome the fact that selection into liquidation is endoge-
nous. For the instrument to be valid, it must not only strongly affect the probability of conversion
to liquidation, but also, importantly, must satisfy the exclusion restriction condition. Specifically,
it is required that judge assignment only affects the outcomes of interest (e.g. whether a plant
location is occupied five years after bankruptcy filing) via its impact on the probability that a case
43The proportion of firms that would not be converted even under the strictest judge (“never takers”) can be easilycalculated as 1 - (% always takers) - (% compliers). For example, in the full sample, we estimate that 18% are alwaystakers, 58% are compliers, and 24% are never takers.
23
is converted to liquidation. As evidence in partial support of our identification assumption, Table
5 reports randomization tests that show that our instrument is uncorrelated with a comprehensive
set of firm and plant level characteristics, as well as local and industry conditions.
Column 1 of Table 5 shows that the R2 when we regress �j on the full set of division by year
fixed effects and no other controls is 0.777, suggesting that there is substantial variation in judge
conversion propensities between divisions and over time. In the next column, we explore whether
within a division-year such variation is correlated with bankruptcy case characteristics by adding
controls for plant size and age, firm size, an indicator for whether there were multiple associated
bankruptcy filings, and industry fixed effects. None of these variables is statistically significant and
the R2 is unaffected by their addition. In column 3 we include pre-trends in employment at the
bankrupt plant for 3 years prior to the bankruptcy filing. These employment figures are calculated
by using the address matching algorithm discussed in Section IV.A to calculate total employment
at a location prior to bankruptcy even if it was not owned by the bankrupt firm. As can be seen,
we find no evidence that judge leniency is correlated with these pre-trends. We provide further
evidence for the lack of pre-trends in Figure 5, which displays the reduced-form results in the years
around the bankruptcy filing. In all cases, �j is uncorrelated with utilization in the years prior to
the bankruptcy filing.
The next columns in Table 5 explore whether local market conditions are correlated with the
instrument. In columns 4, 5, and 6 we separately add dummy variables indicating if a plant was in a
county with above-median market thickness (as defined in Section III), share of small business loans
(also defined in Section III), or three-year cumulative employment growth prior to the bankruptcy.
In column 7 we add all three measures together. In none of the specifications are any of these
measures statistically significant. In column 8 we also add additional variables that capture local
economic activity and industry conditions such as the number of plants in the county and industry,
payroll per employee in the county and industry, and three-year employment growth in an industry.
Once again, all controls are insignificant and the overall R2 remains basically unchanged. The
evidence in Table 5 suggests that there is indeed random assignment of judges to bankruptcy filings
within court divisions, thus alleviating the concern that �j might be related to other factors that
might influence future plant outcomes.
The exclusion restriction assumption might still be violated if judge leniency affects plant out-
comes through channels other than the bankruptcy regime. At this point, it is important to clarify
the definition of the liquidation treatment in our setting. It may be the case that in the economy,
the motion of Chapter 7 conversion is systematically correlated with other motions, or may system-
atically be approved by judges with particular characteristics. If this is how firms are liquidated
24
in the economy, then naturally this is also the liquidation treatment in our setting. We cannot
separate the law from the way it is implemented. In that case, the liquidation treatment should be
viewed more broadly than just the motion to convert to Chapter 7, but rather as the package of
motions and judge characteristics that typically lead to conversion, and the results should be inter-
preted accordingly. Below, we attempt to explore the extent to which this broader interpretation is
warranted.
We first estimate reduced-form regressions which directly relate judge leniency, �j , to plant
outcomes:
Yp,i,t+k = ↵+ � · �j + �Xp,i,t + �d,t + µk + ✏p,i,t+k.
These regressions, reported in Table A.4 in the Appendix, illustrate a strong relationship between
the instrument, �j , and Yp,i,t+k for all of our outcome variables. Arguably, this is because judge
leniency leads to liquidation, which subsequently affects asset reallocation. However, if �j is sys-
tematically correlated with judge skill or other judge attributes that affect asset allocation outside
of the bankruptcy regime, then �j should affect Yp,i,t+k also when limiting the sample only to firms
that remain in reorganization, or only to firms that are liquidated. As reported in Table A.5 in the
Appendix, when we run reduced-form regressions on these two subsets of firms we find no significant
relationship between the instrument and plant outcomes, however. In column 7 of Table A.5, we
also find that within Chapter 11 reorganization, �j is uncorrelated with bankruptcy refiling rates, a
proxy for bankruptcy resolution success which may depend on judge skill. This suggests that other
judge characteristics or tendencies that may be correlated with �j do not affect plant outcomes
outside of the bankruptcy regime.
In further support of this, we also examine whether judges have a large effect on bankruptcy
cases before making a decision on whether to convert a case or not. Based on a random sample of
200 cases, we calculate that the median time between the bankruptcy filing and the selection of the
bankruptcy regime (either liquidation or reorganization) is only 4 months. Further, we find that
typically no significant motions are passed in the case prior to a ruling on a motion to convert the
case. Lastly, we also note that Chang and Schoar (2013), who use detailed data on court motions
to perform a principal component analysis on a set of the most important rulings of a bankruptcy
judge in an effort to identify pro-debtor judges, find that the motion to convert a case receives by
far the lowest weight in the first principal component. This suggests that the decision to convert
may be mostly unrelated to a judge’s overall pro-debtor or pro-creditor bias, as opposed to other
motions. Hence, while we cannot fully reject the broader interpretation of the liquidation treatment,
we find no evidence for its existence in affecting asset allocation and utilization.
25
VI. Results
A. Main Results
We first focus on how liquidation affects reallocation and utilization in the full sample by testing
its impact on four main outcome variables. Continues is an indicator variable equal to one if the
plant is active (has positive payroll) and continues to be occupied by the original bankrupt firm
five years after the bankruptcy filing. The purpose of this variable is to explore the extent to which
liquidation forces more discontinuation than reorganization. Related to the Continues variable is
Holding Time, defined as the number of years until a plant is no longer occupied by the original
bankrupt firm (either reallocated to a new user, or reallocated to vacancy).44 As discussed in the
theoretical framework in Section II, this variable conveys the extent to which liquidation speeds up
the discontinuation of a plant versus reorganization. Finally, we examine two measures of utilization
of real estate assets, regardless of who the occupant is. Occupied is an indicator equal to one if the
asset is occupied five years after the bankruptcy filing. Ln(average employment) is defined as the
average employment at a specific location over the five years after the bankruptcy filing. Because
vacant establishments by definition have zero employment and payrolls, this employment measure
accounts for any interim years in which a plant is not occupied, even if it is occupied in year five.
Further, it has the advantage of accounting for the intensive margin of employment as well as the
extensive margin, since it reflects plants that are reallocated but have fewer employees. For both
measures of utilization, the geographical linkages discussed in Section IV.A allow us to account for
reallocation of assets to new users.
Table 6 shows both OLS and 2SLS estimates of the impact of liquidation on these plant out-
comes.45 These regressions include all controls in column 3 of Table 3, including industry and
division-by-year fixed effects. Regular OLS results, which do not account for selection, show that
liquidation is associated with a 30% decrease in the likelihood of continuation five years after the
bankruptcy filing. The 2SLS estimates in column 2, show that exogenously converting a firm to
liquidation reduces the probability of continuation, with a magnitue of 32.4%. In addition, columns
3 and 4 show that asset holding time is on average 1.7 years shorter in liquidation. These results are
somewhat mechanical, since liquidation forces discontinuation while reorganization does not, and
thus it is not surprising that it causes more discontinuation and speeds up the time until the asset
is no longer used by the bankrupt firm. However, these findings are useful for three reasons. First,
44We cap Holding Time at 5 years to match the horizon in other variables. However, the results are unchanged ifwe set it to 7 or 10 years for these plants.
45As noted previously, observations are weighted by the inverse of the number of establishments in the bankruptfirm to avoid overweighting a few large bankruptcy cases. However, we find essentially identical results in unweightedOLS regressions.
26
they validate that liquidation indeed causes increased shut down. Second, they demonstrate that
reorganization leads to a significant amount of discontinuation as well. Indeed, given that liquida-
tion forces discontinuation, it is somewhat surprising that only 32.4% more firms are discontinued
in liquidation than reorganization. Third, the holding time result relates directly to the theoretical
framework, showing that forced sales do indeed speed up discontinuation as predicted by the model.
We return to this prediction in the heterogeneity analysis below.
The main findings of the paper relate to the utilization of the location regardless of the owner,
as captured by occupied and ln(average employment) in columns 5-8 of Table 6. Across all spec-
ifications, we find that liquidation leads to a significant decline in asset utilization 5 years after
the bankruptcy filing. 2SLS estimates show that liquidation reduces occupancy rates by 17.4%, an
effect that is both statistically and economically significant.46 This estimate is roughly half the size
of the 32.4% decline in plant continuation, demonstrating that reallocation to new users closes some
of the gap between liquidation and reorganization, but not entirely.
The magnitude of the decline is even larger when measuring by employment, estimated at 34%
in column 8.47 This suggests that not only does liquidation reduce occupancy rates on average
(the extensive margin), but it also reduces employment, which proxies for the extent to which an
occupied asset is used.48
While Table 6 focuses on outcomes in year 5 after the bankruptcy, the effect of liquidation on
occupancy and employment is apparent quite quickly after bankruptcy. This can be seen in Figure
5, which shows coefficients from reduced-form regressions over a time horizon from three years prior
to five years after the bankruptcy filing. As shown in Panels (a) and (b), liquidation leads to a
decline in both occupancy and employment in the first year after banrkuptcy, and this drop is quite
stable over time. However, these results mask considerable heterogeneity across markets, which we
discuss in the next section.49
Taken together, the results show that bankruptcy regimes importantly affect asset allocation
and subsequent utilization. In liquidation, plants are more likely to be discontinued, as expected,46It is also interesting to note the gap between the OLS and IV estimates, which capture the selection into
treatment. While there is clearly selection into Chapter 7 liquidation, how this selection might bias OLS estimates isex ante unclear. On one hand, it is likely that poorly-performing firms will be more likely to be converted, and theirassets are less likely to be reallocated, which would bias OLS coefficients downwards. On the other hand, firms withassets that will be easily redeployed may be more likely to move to liquidation, which would bias OLS coefficientsupwards. Results in Table 6 suggest that to a large extent these two effect balance each other out, so that OLSestimates are similar to 2SLS.
47Since these are log-linear models with the independent variable of interest, Liquidationp,i,t, being a dummyvariable, the estimated impact of moving from reorganization to liquidation is 100 [exp(�)� 1].
48In fact, focusing on manufacturing firms only, we find that liquidaiton reduces productivity as reported in TableA.12 in the Appendix. However, this analysis is fairly suggestive due to data limitations, as we discuss in Section Ain the Appendix.
49Appendix Table A.7, which displays dynamic 2SLS coefficient estimates instead of the reduced-form coefficents,also shows that liquidation leads to lower overall wages at the plant.
27
but these assets are not fully reallocated, such that assets in liquidation exhibit lower utilization
relative to reorganization, as measured by either occupancy or employment.
B. The Role of Market Thickness and Access to Finance
The results presented so far show that liquidation causes significantly faster discontinuation and
lower asset utilization five years after the bankruptcy filing. In this section, we explore how the
gap between liquidation and reorganization is related to the market in which bankruptcy occurs. In
particular, we focus on two local market characteristics (described in Section III above) that theory
predicts affect asset reallocation: market thickness and access to finance.
In Table 7, we split the sample based on the market thickness measure, in which we define
“thick” industry-county-year triplets as those having above-median Thicknessict, and then run our
IV specifications separately for plants in thick and thin markets. We also display results in which we
interact liquidation with this thick market dummy, and correspondingly we interact share converted
with thick market in the first stage so that we have two endogenous variables and two instruments.
To allow full flexibility, we interact all other covariates with thick market as well. With this setup,
the coefficient on liquidation*thick market tests whether the effect of liquidation is significantly
different in thick markets versus thin markets.50
The first three columns of Panel A show that in both thick and thin markets liquidation reduces
the probability that a plant will continue with the original bankrupt firm by a similar amount.
However, columns 4-6 show that this discontinuation is faster in liquidation in thin asset markets.
In thick markets, reorganized plants are held by the original owner for 1.4 more years than liquidated
plants, while in thin markets the gap is 29% larger at 1.8 years.51 This is consistent with the dynamic
search model in Section II, which predicts that liquidation leads to a quicker reallocation of assets
in particular in thin markets. In these markets, reorganized firms can be patient as they wait to
find a new user, while liquidated plants are forced to try to locate new users despite high search
costs.
Panel B of Table 7 shows that the differences in asset utilization between thick and thin mar-
kets are stark. In column 1, we find that asset reallocation in thick markets allows the occupancy
rate of liquidated plants to be similar to that of reorganized plants, despite the high level of dis-
continuation in liquidation shown in Panel A. Thus, the null hypothesis of no difference between
50Note that we do not claim that plants are exogenously distributed across thick or thin markets. By running theregressions on separate sub-samples (or fully interacted), we compare thick-market firms that are randomly assignedto “liquidating judges” to those that are assigned to “reorganizing judges,” and similarly we compare thin-market firmsthat are randomly liquidated to those that are not. Thus within each regression the estimates can still be interpretedas causal, and the comparison across regressions sheds light on which markets are driving the overall effects.
51While the difference in holding times is economically important, the gap is not statistically significant.
28
the two bankruptcy regimes is not rejected, as the market fully absorbs the increased numbers of
discontinued plants in liquidation. Meanwhile, column 2 shows that occupancy rates for liquidated
plants are 32.4% lower in thin asset markets, relative to plants that are reorganized in thin mar-
kets. Hence, in contrast to thick markets, liquidated plants do not seem to reallocate to new uses
at higher rates than reorganized plants.52 Column 3 shows that the difference between the two
markets is statistically significant. Similarly, in comparing columns 4 and 5 we find that in thick
markets liquidation does not have a significant effect on average employment (indeed, the coefficient
estimate is even positive), but in thin markets liquidation reduces employment by 54.6%. Again,
this difference between the two markets is statistically different. Overall, the effect of liquidation on
asset utilization is entirely concentrated in thin markets, while reallocation in thick markets results
in liquidation having no impact on utilization. These results are consistent with the theoretical
framework in Section II that highlights the implications of search frictions and thin markets on
asset reallocation (Williamson (1988); Gavazza (2011)).53
Similar to market thickness, access to finance can affect asset allocation and utilization in
bankruptcy regimes by limiting the set of potential users (Shleifer and Vishny (1992)). In Ta-
ble 8 we present regression results similar to Table 7, except here we split our sample based on
access to capital in the local market. We proxy for access to finance by measuring for each county
the share of loans given to small businesses, defined as firms with $1 million or less in annual gross
revenue.54 In all cases, we find similar results as those for market thickness. Specifically, in markets
with low access to finance liquidation reduces holding times by 2 years, while in areas with high
access to finance the gap is 1.4 years. Further, we see no decline in occupancy or employment in
markets with high access to finance, but 45% lower occupancy and 69% lower employment in areas
with low access to capital. This supports the theory that search costs, and access to capital in
particular, are a key determinant in the ability to reallocate bankrupt assets.55
52Indeed, the coefficient estimate of liquidation’s impact on continuation in thin markets (Column 2, -32.1%)is almost identical to its impact on occupancy (Column 4, -32.4%). This does not mean, however, that there isno reallocation of liquidated plants in thin markets. Rather, it shows that the reallocation of assets increases theoccupancy of both reorganized and liquidated plants at similar rates in thin markets.
53These results are robust to using an alternative measure of market thickness, that is, local commercial real estatetransactions per capita. This measure aims to capture the liquidity of the local commercial real estate market. Weconstruct this measure using the CoreLogic dataset by dividing the total number of real estate transactions in acounty by the county population. The results are discussed in detail in the Appendix, and reported in Panel A ofTable A.11 in the Appendix.
54Loan data comes from the Community Reinvestment Act (CRA) disclosure data and is only available beginningin 1996, which removes about 30,000 plants from our sample that filed for bankruptcy prior to 1996. Note that theCRA data is based on the location of the loan recipient rather than the location of the bank, and thus the bank isnot necessarily located in the same county. However, the 2003 Survey of Small Business Finances shows that bankloan markets tend to be quite local, as over 70% of firms borrow from banks located less than 20 miles away.
55We find similar results when using an alternative measure of local access to finance, that is, the share of bankdeposits in a county held small banks. This variable stems from the idea that small, local banks are the principleproviders of capital for small firms (Petersen and Rajan (1994)). Hence, a higher concentration of deposits in local
29
In Panels (c) - (f) of Figure 5, we examine how the effect of liquidation evolves dynamically
in markets with high and low search costs. These figures show the coefficients from reduced form
regressions where we interact the instrument with dummy variables of high and low search costs for
market thickness and access to finance respectively. Importantly, we find no pre-trends in occupancy
or employment for any of these sample splits. In the post-bankruptcy period, Panels (c) and (d)
show that liquidation in thick markets causes an initial drop in utilization in year 1, but that this gap
disappears by year 2, consistent with low search costs enabling asset reallocation after the liquidated
firm is shut down. However, the drop in utilization caused by liquidation slightly increases over
time in thin markets, where search costs are high. A similar pattern emerges in Panels (e) and (f),
where we split by areas with high and low access to finance.56
Whether we proxy for search costs with market thickness or access to finance, we find support
for the theoretical predictions outlined above. Importantly, market thickness is uncorrelated with
access to finance, as shown in Appendix Table A.10, suggesting that each channel is separate and
exerts a significant effect individually. These results are robust when using alternative measures of
market thickness and access to capital, as described in Appendix Section A. We have also performed
several additional robustness tests in unreported results. For example, we find essentially identical
results if we split by market thickness within industry-year, showing that the results are not driven
by certain industries or by time-series variation, but rather by geographic variation within industry.
The results are not driven by rural counties with few potential buyers. Dropping all counties with
less than 20,000 employees (the 10th percentile in our sample) does not affect the results. Further,
splitting the sample by a measure of industry agglomeration developed by Ellison and Glaeser
(1997), which is similar to our Thicknessict measure but explicitly adjusts for county size, shows
similar results. In addition, scaling access to capital on a per capita basis, rather than market share,
does not affect the results.
C. External Validity and Economic Importance
While our results show that liquidation has a large impact on the long-term utilization of bankrupt
locations, the aggregate economic importance of these findings could be small if only the smallest
firms are liquidated, or few firms are at the margin between the two bankruptcy procedures. This
section discusses the external validity of our results and their overall economic importance.
In Section V.C above, we have already shown that over 40% of firms with less than 100 employees
small banks is likely to provide higher access to capital to small businesses. We discuss the construction of themeasure in the Appendix, and report the results in Panel C of Table A.11 in the Appendix.
56For robustness, Appendix Figure A.1 displays similar results using a binary version of our instrument in reduced-form regressions with no other controls or fixed effects.
30
are converted to liquidation, and even among firms with 100-1,000 employees 29% are liquidated.
Further, we estimate that moving from the most lenient to the most strict judge would shift between
49% and 88% of all firms with fewer than 1,000 employees to liquidation, suggesting that the
majority of firms with less than 1,000 employees are marginal in the sense that whether they end up
in reorganization or liquidation depends on judge assignment. This corroborates Bris et al. (2006),
who find that there is considerable overlap in the size of Chapter 7 and Chapter 11 firms, and that
“a good number of firms could have chosen either procedure.”
Further, firms with less than 1,000 employees constitute a significant portion of the U.S. economy.
Using data from the full LBD, we find that firms with less than 1,000 employees compose 99.8%
of all firms between 1992 and 2005 (our sample period). These firms employ 55% of all workers
and, more importantly for this study, occupy 85% of all business locations in the U.S. Thus, a large
portion of the assets in the economy are held by firms that could marginally be placed in liquidation
or reorganization.
A second important point is that many firms in the U.S. file directly for Chapter 7. According
to U.S. Court filing statistics, direct-to-Chapter-7 filings account for 72% of all non-farm business
bankruptcies, while Chapter 11 filings make up the remaining 28%.57 For identification purposes,
our sample is composed only of firms that file for Chapter 11 (of which 40% are subsequently
converted to Chapter 7), but to the extent that all firms in Chapter 7 face search costs when
liquidated, our results can be interpreted more broadly.
VII. Efficiency Discussion
While the empirical results in the paper demonstrate that liquidation leads to lower utilization in
thin markets and markets with low access to finance, the question remains whether this gap in
utilization is caused by inefficient liquidation or inefficient continuation of reorganized firms. On
the one hand, liquidation may lead to under-utilization if reallocation is hampered by high search
costs due to thin markets (Williamson (1988); Gavazza (2011)), or financial constraints (Shleifer
and Vishny (1992)). Alternatively, it might be the case that assets are over-utilized in thin markets
due to agency costs that lead to the inefficient continuation of reorganized firms (Franks and Torous
(1989); Gertner and Scharfstein (1991); Bolton and Scharfstein (1996); Hotchkiss (1995)). In this
section we conduct several tests, both theoretical and empirical, to explore whether the results arise
through inefficient reorganization or inefficient liquidation. Importantly, this discussion is focused
57In Appendix Table A.1 we present summary statistics for a sample of firms that filed directly for Chapter 7that we matched to the LBD. The average Chapter 7 firm employs 20.4 workers and occupies 1.4 locations, which issignificantly smaller than the average Chapter 11 firm, but not negligible in the aggregate.
31
exclusively on the efficiency of the ex post asset allocation of each bankruptcy procedure. Other
costs and benefits of liquidation and reorganization, including legal fees, creditor recoveries, and
worker outcomes, play an important role in the overall welfare implications of each bankruptcy
regime, but are not considered here.
A. Agency Costs and Asset Allocation
Although we cannot completely determine whether liquidation is associated with inefficiently low
levels of utilization or reorganization leads to inefficiently high levels of utilization, we have some
evidence against the latter. In particular, the main argument for liquidation being efficient is that
it leads to an improved allocation of assets once a match is found, even if this comes at a cost of
leaving an asset vacant for a period of time. Meanwhile, assets in reorganization might be prevented
from reallocating to better uses due to agency costs. If this is the case, then we should see that the
utilization of liquidated assets is higher than that of reorganized assets once a match has been found
and the location has been reallocated. Following this logic, if agency costs are high we would expect
the utilization of liquidated assets to be higher in thick asset markets where reallocation happens
relatively quickly for these assets. However, this is inconsistent with the empirical findings.
The dynamic search model described in Section II, and developed formally in Appendix B,
formalizes this intuition. The model incorporates both search frictions and agency costs, and also
includes the time dynamics that capture the opportunity cost of waiting for future potential users.
As illustrated in Figure 1, our empirical evidence fully supports the model predictions when agency
problems are modest.
Figure 2 illustrates how the model’s predictions change when agency costs become significant.58
Panels (a) and (b) show that, when agency costs are significant, only reorganized sellers with
particularly low productivity become active sellers, due to their desire to retain the asset. This
affects the holding time as illustrated in panels (c) and (d), where the holding time increases for every
level of market thickness when agency costs are high.Therefore, increasing agency costs increases
the fraction of low-productivity reorganized sellers that continue to hold the assets inefficiently.59
Most strikingly, significant agency costs lead to different predictions with respect to asset uti-
lization and productivity of asset users over time. As illustrated by Panel (f) of Figure 2, when
agency costs are significant, the utilization of the reorganized assets is lower than liquidation, even
in thin markets, and as the market becomes thicker, the gap between liquidation and reorganization
58We formally derive the model, and discuss the construction of the numerical illustration in Section B of theAppendix.
59Note that in panels (c) and (d) we plot the log of holding time to easily compare high and low agency costsusing the same y-axis scale.
32
increases and there is no crossing-point between the two curves. This is the case since liquidated
sellers reallocate the asset to new and more productive buyers while reorganized sellers are unwilling
to do so, even when their productivity is particularly low.
Due to agency costs, reorganized sellers are insulated from the market and thus less responsive to
reductions in search costs, which leads to the increase in the gap in utilization. This is inconsistent
with the empirical evidence, however, as we find that on average reorganization leads to higher
utilization in thin markets, and the gap is decreasing in market thickness such that empirically we
find no significant differences between liquidated and reorganized sellers in thick markets, equivalent
to the cross-point between the utilization curves. These empirical results are consistent with low
agency costs as illustrated in Panel (e), in which there is a crossing point when markets become
thicker. A similar picture arises with the average productivity of the users of the assets, as illustrated
in Panels (g) and (h).
The model highlights that our empirical findings are inconsistent with large agency costs pre-
venting assets from reallocating in reorganization. We should note, however, that this does not
suggest that agency costs are not important in bankruptcy. Indeed, agency costs are likely to be
significant in the very largest bankruptcy cases with complex capital structures. But liquidation in
these large firms is uncommon, as shown in Section V.C. Instead, our results, which focus on the
large set of firms that are on the margin between the two bankruptcy regimes, suggest that search
frictions loom larger than agency costs. Therefore, inefficient allocation of assets is likely to occur
in thin markets when firms are liquidated and experience high search costs.
B. Opportunity Cost and Local Vacancies
From an empirical standpoint, understanding whether an asset is efficiently vacant or under-utilized
is challenging because of the difficulty in measuring its opportunity cost. However, it is important
to note that the random assignment of judges in the empirical strategy effectively allows us to
compare plants with similar opportunity costs. Specifically, since the assets’ opportunity cost is
orthogonal to the judge leniency to convert cases to Chapter 7 (as all other characteristics), it can
be viewed as if the liquidation treatment occurs while holding opportunity cost constant. Therefore,
the lower utilization caused by liquidation relative to reorganization is arguably independent of the
opportunity cost of the asset.
However, it might still be the case that liquidation causes lower utilization only in areas where
the opportunity cost of utilizing the assets is low, and that this is driving our overall estimates. For
example, liquidation could lead to high vacancy rates but have little impact on overall efficiency if
local availability of space is high. This suggests a natural proxy for the oppotunity cost of vacancy:
33
the utilization of non-bankrupt local establishments. In particular, we posit that the opportunity
cost of leaving a location vacant depends on the overall vacancy rate of the local area.60 In areas
with low vacancy rates, the opportunity cost of leaving a location vacant should be high since
there is relatively high demand for real estate. Meanwhile, areas with high vacancy will have low
opportunity costs of vacancy since firms may have many alternative locations to move into. Under
this assumption, if we find that liquidation causes a drop in utilization even in areas with low
vacancy (high opportunity cost) then it provides further evidence consistent with the inefficiency of
liquidation.
In Figure 3 we compare the utilization of bankrupt locations relative to average local utilization
of non-bankrupt plants, which proxies for the opportunity cost of vacancy. As can be seen, bankrupt
assets are less likely to be occupied and utilized relative to average local plants for all five years
after the bankruptcy. More importantly, liquidation is associated with significantly lower utilization
relative to reorganization. This evidence suggests inefficient use of the assets relative to the local
characteristics.61
We directly test whether liquidation leads to vacancy only when opportunity costs are low in
Table 9. In column 1 of Panel A, we show that our main 2SLS specification is unaffected when we
control for an indicator equal to one if the bankrupt plant is in a county with high (above-median)
benchmark utilization. Interestingly, this control strongly predicts the utilization of the bankrupt
firm’s asset. However, the coefficient of liquidation remains unchanged, which rules out an omitted
variables bias in which the negative effect of liquidation stems from plants being left vacant in areas
that already have low occupancy. Column 2 shows that the results with respect to market thickness
are robust and unaffected by the control for local benchmark occupancy as well.
Even more telling is column 3, where we also include an interaction between the liquidation
dummy and the high benchmark occupancy indicator.62 While the direct effect of high benchmark
occupancy is statistically significant and predicts whether the bankrupt plant is utilized, we find that
the coefficient on this interaction term is insignificant and near zero. This means that liquidation
does not have a differential effect in areas with high benchmark occupancy, when the opportunity
cost of vacancy is arguably low. This is important, as it shows that liquidation causes a decline in
60Grenadier (1995, 1996) finds that the level of equilibrium vacancy rates is predominately determined by localfactors. Moreover, he illustrates a significant persistence in local vacancy rates in commercial real estate.
61We do not have an instrument for entering bankruptcy, and so this comparison with benchmark plants is onlysuggestive because it does not account for selection effects. For example, bankrupt buildings could remain vacant athigher rates because they are worse in some way. To the extent that these locations enter bankruptcy because thefirms (and not the buildings) are bad, this is less of a concern. That is, as long as a building can be reallocated to amore-productive user, we should see utilization of these locations approach that of the overall economy.
62As before, when interaction terms are included in this second stage regression, we include the interaction of theshare converted instrument and that variable in the first stage.
34
utilization even in areas with high opportunity costs of vacancy. Columns 5 and 6 show further that
the results regarding access to capital and liquidation are not dependent on benchmark occupancy
either. Panel B demonstrates that all of the results with respect to employment are unaffected when
controlling for benchmark utilization as well.63
Taken together, these results provide consistent evidence that our results are not driven by
high-vacancy areas where the opportunity cost of leaving a location vacant are likely low. Instead,
the evidence is consistent with the interpretation that liquidation causes vacancy even when the
opportunity cost of doing so is high, making it less likely that this is an efficient outcome in thin
markets. Of course, because we cannot directly measure asset efficiency, these results are not fully
conclusive. However, they point towards the conclusion that liquidation leads to inefficient asset
allocation and utilization in areas with high search costs.
VIII. Conclusion
How do institutions affect the allocation of assets in the economy? In this work we explore the role
of the bankruptcy system in the allocation of the distressed firms’ assets. In particular, we explore
how liquidation and reorganization affect the allocation and subsequent utilization of the real estate
assets used by bankrupt firms. We exploit the random assignment of judges to bankruptcy cases
and variations in judges’ interpretation of the law to instrument for the endogenous conversion
of Chapter 11 filers into Chapter 7 liquidation. We create unique geographical linkages from the
Census LBD database that allow us to track real estate utilization over time.
We find that liquidation leads to a significant reduction in the utilization of real estate assets
on average, and this effect persists in the five years after the bankruptcy filing. These effects are
fully concentrated in areas with high search costs such as thin asset markets where there are few
potential users for bankrupt assets, and in areas with low access to finance. In contrast, in markets
with low search frictions we find no differential effect of bankruptcy institutions on asset utilization.
The results highlight the importance of local asset market frictions in the allocation of assets of
firms in bankruptcy.
63All of the results also go through if we use as the benchmark only plants in the same county and 3-digit NAICSas the bankrupt plant. Further, the results are unchanged if we use the continuous benchmark occupancy rate ratherthan the above-median indicator as a control.
35
REFERENCES
Aghion, Philippe, Oliver Hart, and John Moore, “The economics of bankruptcy reform,”Journal of Law, Economics and Organization, 1992, 8 (3), 523–546.
Ayotte, Kenneth, “Leases and Executory Contracts in Chapter 11,” Journal of Empirical LegalStudies, 2015, 12 (4), 637–663.
Baird, Douglas G, “The uneasy case for corporate reorganizations,” The Journal of Legal Studies,1986, 15 (1), 127–147.
, “Revisiting auctions in Chapter 11,” Journal of Law and Economics, 1993, 36 (1), 633–653.
Bartelsman, Eric, John Haltiwanger, and Stefano Scarpetta, “Cross-country differences inproductivity: The role of allocation and selection,” The American Economic Review, 2013, 103(1), 305–334.
Bebchuk, Lucian Ayre and Howard F Chang, “Bargaining and the division of value in corpo-rate reorganization,” Journal of Law, Economics, & Organization, 1992, 8 (2), 253–279.
Benmelech, Efraim and Nittai K Bergman, “Bankruptcy and the collateral channel,” TheJournal of Finance, 2011, 66 (2), 337–378.
Bolton, Patrick and David S. Scharfstein, “Optimal debt structure and the number of credi-tors,” The Journal of Political Economy, 1996, 104 (1), 1–25.
Bradley, Michael and Michael Rosenzweig, “The untenable case for Chapter 11,” Yale LawJournal, 1992, 101 (5), 1043–1095.
Bris, Arturo, Ivo Welch, and Ning Zhu, “The costs of bankruptcy: Chapter 7 liquidationversus Chapter 11 reorganization,” The Journal of Finance, 2006, 61 (3), 1253–1303.
Campbell, John Y., Stefano Giglio, and Parag Pathak, “Forced Sales and House Prices,”American Economic Review, 2011, 101 (5), 2108–2131.
Chang, Tom and Antoinette Schoar, “Judge specific differences in chapter 11 and firm out-comes,” 2013.
Dahl, Gordon B., Andreas Ravndal Kostøl, and Magne Mogstad, “Family Welfare Cul-tures,” The Quarterly Journal of Economics, 2014, 129 (4), 1711–1752.
Davis, Steve J. and John Haltiwanger, “Gross job creation, gross job destruction and employ-ment reallocation,” The Quarterly Journal of Economics, 1992, 107 (3), 819–863.
Davydenko, Sergei A and Julian R Franks, “Do bankruptcy codes matter? A study of defaultsin France, Germany, and the UK,” The Journal of Finance, 2008, 63 (2), 565–608.
Djankov, Simeon, Oliver Hart, Caralee McLiesh, and Andrei Shleifer, “Debt Enforcementaround the World,” Journal of Political Economy, 2008, 116 (6).
Dobbie, Will and Jae Song, “Debt Relief and Debtor Outcomes: Measuring the Effects ofConsumer Bankruptcy Protection,” American Economic Review, 2015, 105, 1272–1311.
36
Doyle Jr, Joseph J., “Child protection and child outcomes: Measuring the effects of foster care,”The American Economic Review, 2007, pp. 1583–1610.
Doyle Jr., Joseph J., “Child protection and adult crime: Using investigator assignment to estimatecausal effects of foster care,” Journal of Political Economy, 2008, 116 (4), 746–770.
Eckbo, Espen B and Karin Thorburn, “Automatic bankruptcy auctions and fire-sales,” Journalof Financial Economics, 2008, 89, 404–422.
Eisfeldt, Andrea L and Adriano A Rampini, “Capital reallocation and liquidity,” Journal ofMonetary Economics, 2006, 53 (3), 369–399.
Ellison, Glenn and Edward L Glaeser, “Geographic concentration in U.S. manufacturing in-dustries: A dartboard approach,” Journal of Political Economy, 1997, 105 (5), 889–927.
, , and William R Kerr, “What Causes Industry Agglomeration? Evidence from Coagglom-eration Patterns,” American Economic Review, 2010, 100, 1195–1213.
Franks, Julian R. and Walter N. Torous, “An empirical investigation of U.S. firms in reorga-nization,” The Journal of Finance, 1989, 44 (3), 747–770.
Galasso, Alberto and Mark Schankerman, “Patents and cumulative innovation: Causal evi-dence from the courts,” The Quarterly Journal of Economics, 2015, 130 (1), 317–369.
Gavazza, Alessandro, “The role of trading frictions in real asset markets,” The American Eco-nomic Review, 2011, 101 (4), 1106–1143.
Gertner, Robert and David Scharfstein, “A theory of workouts and the effects of reorganizationlaw*,” The Journal of Finance, 1991, 46 (4), 1189–1222.
Greenstone, Michael, Alexandre Mas, and Hoai-Luu Nguyen, “Do credit market shocksaffect the real economy? Quasi-experimental evidence from the Great Recession and normaleconomic times,” Technical Report, National Bureau of Economic Research 2014.
Grenadier, Steven R, “Local and national determinants of office vacancies,” Journal of UrbanEconomics, 1995, 37 (1), 57–71.
, “The strategic exercise of options: Development cascades and overbuilding in real estate mar-kets,” The Journal of Finance, 1996, 51 (5), 1653–1679.
Hart, Oliver, “Different approaches to bankruptcy,” Technical Report, National Bureau of Eco-nomic Research 2000. Working Paper 7921.
He, Zhiguo and Gregor Matvos, “Debt and creative destruction: Why could subsidizing corpo-rate debt be optimal?,” Management Science, 2015, 62 (2), 303–325.
Hotchkiss, Edith S., “Postbankruptcy performance and management turnover,” The Journal ofFinance, 1995, 50 (1), 3–21.
Hsieh, Chang-Tai and Peter J Klenow, “Misallocation and manufacturing TFP in China andIndia,” The Quarterly Journal of Economics, 2009, 124 (4), 1403–1448.
Imbens, Guido W and Joshua D Angrist, “Identification and estimation of local averagetreatment effects,” Econometrica, 1994, 62 (2), 467–476.
37
Ivashina, Victoria, Benjamin C. Iverson, and David C. Smith, “The ownership and tradingof debt claims in Chapter 11 restructurings,” Journal of Financial Economics, Forthcoming, 2015.
Iverson, Benjamin C., “Get in line: Chapter 11 restructuring in crowded bankruptcy courts,”Available at SSRN 2156045, 2015.
Jarmin, Ron S and Javier Miranda, “The longitudinal business database,” Available at SSRN:http://ssrn.com/abstract=1509273, 2002.
Jensen, Michael C and William H Meckling, “Theory of the firm: Managerial behavior, agencycosts and ownership structure,” Journal of financial economics, 1976, 3 (4), 305–360.
Kling, Jeffrey R., “Incarceration Length, Employment, and Earnings,” American Economic Re-view, 2006, 96 (3), 863–876.
LoPucki, Lynn M and William C Whitford, “Patterns in the bankruptcy reorganization oflarge publicly held companies,” Cornell Law Review, 1993, 78 (4), 597–618.
Maestas, Nicole, Kathleen J Mullen, and Alexander Strand, “Does disability insurancereceipt discourage work? Using examiner assignment to estimate causal effects of SSDI receipt,”The American Economic Review, 2013, 103 (5), 1797–1829.
, , and , “Does disability insurance receipt discourage work? Using examiner assignment toestimate causal effects of SSDI receipt,” American Economic Review, 2013, 103 (5), 1797–1829.
Maksimovic, Vojislav and Gordon Phillips, “Asset efficiency and reallocation decisions ofbankrupt firms,” The Journal of Finance, 1998, 53 (5), 1495–1532.
Mortensen, Dale and Randall Wright, “Competitive pricing and efficiency in search equilib-rium,” International Economic Review, 2002, 43 (1), 1–20.
Ottonello, Pablo, “Capital Unemployment, Financial Shocks, and Investment Slumps,” 2014.
Petersen, Mitchell A and Raghuram G Rajan, “The benefits of lending relationships: Evidencefrom small business data,” The Journal of Finance, 1994, 49 (1), 3–37.
Pulvino, Todd C, “Do asset fire sales exist? An empirical investigation of commercial aircrafttransactions,” The Journal of Finance, 1998, 53 (3), 939–978.
, “Effects of bankruptcy court protection on asset sales,” Journal of Financial Economics, 1999,52 (2), 151–186.
Ramey, Valerie A. and Matthew D. Shapiro, “Displaced capital: A study of aerospace plantclosings,” Journal of Political Economy, 2001, 109 (5), 958–992.
Shleifer, Andrei and Robert Vishny, “Fire Sales in Finance and Macroeconomics,” Journal ofEconomic Perspectives, 2011, 25 (1), 29–48.
and Robert W Vishny, “Liquidation values and debt capacity: A market equilibrium ap-proach,” The Journal of Finance, 1992, 47 (4), 1343–1366.
Staiger, Douglas and James H Stock, “Instrumental Variables Regression with Weak Instru-ments,” Econometrica, 1997, 65 (3), 557–586.
38
Strömberg, Per, “Conflicts of interest and market illiquidity in bankruptcy auctions: Theory andtests,” The Journal of Finance, 2000, 55 (6), 2641–2692.
Tella, Rafael Di and Ernesto Schargrodsky, “Criminal recidivism after prison and electronicmonitoring,” Journal of Political Economy, 2013, 121 (1), 28–73.
Williamson, Oliver E, “Corporate finance and corporate governance,” The Journal of Finance,1988, 43 (3), 567–591.
39
Figure 1Model with Bankrupt Sellers
This numerical illustration captures the key insights from a basic search model with respect to searchthresholds, asset holding time, asset utilization, average productivity of asset users for different levels ofmarket thickness, defined in the model as markets with higher mass of assets. See Section B in the Appendixfor details on the construction of the numerical illustration. Black solid lines denote regular buyers, blackdotted lines are regular sellers, red dotted lines are reorganized sellers, and blue solid lines are liquidatedsellers.
200 400 600 800 1000
1416
1820
2224
Mass of assets
Sear
ch T
hres
hold
s
BuyerSellerReorganization
(a) Search Thresholds
200 400 600 800 1000
010
2030
4050
Mass of assets
Expe
cted
Tim
e to
Sal
e
LiquidationReorganizationSeller
(b) Holding Time
200 400 600 800 1000
135
140
145
Mass of assets
Aver
age
Asse
t Util
izat
ion
LiquidationReorganizationSeller
(c) Asset Utilization
200 400 600 800 1000
450
460
470
480
490
500
510
Mass of assets
Aver
age
Asse
t Pro
duct
ivity
LiquidationReorganizationSeller
(d) Asset Productivity
40
Figure 2Agency Costs
This numerical illustration captures the key insights from a basic search model with respect to searchthresholds, asset holding time, asset utilization, average productivity of asset users for different levels ofmarket thickness, defined in the model as markets with higher mass of assets. The plots compare caseswith high level of agency costs (left panel) and low level of agency costs (right panel). See Section B inthe Appendix for details on the construction of the numerical illustration. Black solid lines denote regularbuyers, black dotted lines are regular sellers, red dotted lines are reorganized sellers, and blue solid lines areliquidated sellers.
(a) Search Thresholds - Low
200 400 600 800 1000
510
1520
25Mass of assets
Sear
ch T
hres
hold
s
(b) Search Thresholds - High
200 400 600 800 1000
−10
12
34
Mass of assets
Log
Expe
cted
Tim
e to
Sal
e
(c) Log Holding Time - Low
200 400 600 800 1000
02
46
8
Mass of assets
Log
Expe
cted
Tim
e to
Sal
e
(d) Log Holding Time - High
200 400 600 800 1000
135
140
145
Mass of assets
Aver
age
Asse
t Util
izat
ion
(e) Asset Utilization - Low
200 400 600 800 1000
135
140
145
Mass of assets
Aver
age
Asse
t Util
izat
ion
(f) Asset Utilization - High
200 400 600 800 1000
450
460
470
480
490
500
510
Mass of assets
Aver
age
Asse
t Pro
duct
ivity
(g) Asset Productivity - Low
200 400 600 800 1000
440
460
480
500
Mass of assets
Aver
age
Asse
t Pro
duct
ivity
(h) Asset Productivity - High
41
Figure 3Stylized Facts About Bankruptcy Reallocation
These figures illustrate summary statistics about the reallocation process for liquidated, reorganized,and benchmark establishments. The benchmark is created from a random 5% sample of all plantsin the same county as the bankrupt establishments. Panel A shows the percentage of plants thatcontinue to be operated by the original firms in the 5 years following the bankruptcy filing. PanelB plots the share of establishments that are occupied regardless of the owner, thereby taking intoaccount reallocation to new users. Panel C is similar to Panel B, but instead shows total employmentas a percentage of employment in year 0.
(a) Plant Continuation Probability Over Time
(b) Plant Occupancy Rate Over Time
(c) Plant Employment Over Time
42
Figure 4Non-Parametric First Stage
This figure plots the relationship between the probability of case conversion and our preferred instrument,the share of all other Chapter 11 cases that a judge has converted to Chapter 7, using a non-parametric kernelregression. To be consistent with the regression analysis in the paper, we first residualize the probability ofcase conversion to all control variables in Table 3, including division-year fixed effects. Gray area representsthe 95% confidence interval. For disclosure reasons, we truncate the 5% tails of the distribution.
43
Figure 5Reduced Form Dynamics
This figure displays coefficient estimates of the effect of judge instrument on occupancy rates (leftcolumn) and average employment (right column) both before and after bankrutpcy. In Panels (a)and (b), coefficient estimates � are from reduced-form regressions Yp,i,t+k = ↵+� ·�j+�Xp,i,t+�d,t+µk + ✏p,i,t+k, where �j is judge leniency instrument, and specifications contain the full set of controlvariables in Column 3 of Table 3, including division-by-year fixed effects. In the remaining panels,coefficient estimates displayed are �L and �H , which are derived from reduced-form regressionsYp,i,t+k = ↵+�L ·�j ·Lowp,t+�H ·�j ·Highp,t+�Xp,i,t+�d,t+µk+ ✏p,i,t+k, where the instrument �j
is interacted with dummy variables Lowp,t and Highp,t, which indicate if a plant resides in a countywith low or high market thickness (Panels (c) and (d)) or access to finance (Panels (e) and (f)).All control variables and fixed effects are also interacted with Lowp,t to allow for flexible estimatesacross market types. Standard-error bars, based on clustering at the division-year level, are alsodisplayed.
(a) Occupied-Main Effect (b) Employment-Main Effect
(c) Occupied-By Market Thickness (d) Employment-By Market Thickness
(e) Occupied-By Access to Capital (f) Employment-By Access to Capital
44
Table 1Sample Summary Statistics
Panel A of this table presents summary statistics on the plants and firms in our final sample, both overall andsplit by firms that are reorganized in Chapter 11 and those that are liquidated in Chapter 7. Observationcounts are rounded to the nearest thousand due to disclosure requirements of the U.S. Census. All numbersshown are averages, except for observation counts. Payroll and payroll per employee are in thousands ofnominal U.S. dollars. Market thickness and Share of small business loans are defined in the text. Share ofsmall business loans is only available beginning in 1996, leaving a total of 99,000 plants for these summarystats. Panel B shows the industry distribution in our sample and the percent of firms liquidated in eachindustry. Panel C describes the characteristics of the firms replacing the dead bankrupt plants, distinguishingbetween new firms, existing firms that already had an establishment in the same county, and other existingfirms. In Panel C we also report the percentages of reallocations to the same 2- and 3-digit NAICS industry.
Panel A: Summary StatisticsAll Reorganized Liquidated
Plant-level characteristicsEmployment 35.9 38.0 26.9Total plants 129,000 105,000 24,000Firm-level characteristicsNo. Plants 4.7 6.5 2.2Employment 169.0 245.4 57.9Payroll (000s) 4,507.7 6,819.0 1,146.3Payroll/Employee (000s) 23.7 26.0 20.2Age 9.9 10.7 8.9Number of firms 28,000 17,000 11,000County-level characteristicsMarket thickness 6.4% 6.4% 6.4%Share of small business loans 43.8% 43.7% 43.9%
Panel B: Industry DistributionTotal Plants Total Firms % Liquidated
Agriculture, Mining, and Construction 4,000 3,500 45Manufacturing 7,000 4,500 38Transportation, Utilities & Warehousing 7,000 3,000 43Wholesale & Retail Trade 62,000 6,500 44Services 18,000 5,500 37Accommodation, Food, and Entertainment 16,000 4,000 44Other 15,000 3,500 32
Panel C: New Entrant CharacteristicsAll Reorganized Liquidated
Local vs. non-localNew entrant 32,500 52.0% 23,500 48.0% 9,500 70.4%Local entrant, existing 21,500 34.4% 18,000 36.7% 3,000 22.2%Non-local entrant, existing 8,500 13.6% 7,500 15.3% 1,000 7.4%
Total 62,500 100.0% 49,000 100.0% 13,500 100.0%
Industry transitionsIn same 3-digit NAICS 29,000 46.4% 24,000 49.0% 5,000 37.0%In same 2-digit NAICS 34,500 55.2% 28,500 58.2% 6,000 44.4%
45
Table 2Reallocation Determinants
This table shows results from a regression of a dummy for whether a plant is replaced within 5 yearsfrom bankruptcy filing (conditional on death of original plant) on a set of county and industry (2-digitNAICS) characteristics computed at the year of filing. All county-level and industry-level controls aredummy variables equal to 1 if the county is above-median in the given category. Plant- and firm-levelcontrols identical to those in Table 3 are also included, but are not reported for brevity. In addition, withthe exception of column 4, we include fixed effects for the filing year, as well as for the number of yearsafter plant death in which we looked for a replacement (up to 5 years after filing). The sample includes allestablishments that died within 5 years of filing. Standard errors, clustered at the division by year level,are shown in parentheses. *, **, and *** denote statistical significance at the 10%, 5%, and 1% level,respectively.
Dependent variable: Plant reallocation dummy
(1) (2) (3) (4) (5) (6)
Local Economic ConditionsNo. plants above median 0.029*** 0.034*** 0.029*** 0.028***
(0.005) (0.007) (0.005) (0.005)3-year employment growth above median 0.019*** 0.027*** 0.019*** 0.018***
(0.004) (0.005) (0.004) (0.004)Payroll per employee above median 0.032*** 0.004 0.032*** 0.029***
(0.006) (0.009) (0.007) (0.006)Industry Economic ConditionsNo. plants above median -0.012 -0.017* -0.035* -0.016* 0.025
(0.008) (0.009) (0.019) (0.009) (0.021)3-year employment growth above median 0.026*** 0.005 0.026* 0.006 -0.004
(0.007) (0.009) (0.015) (0.009) (0.009)Payroll per employee above median 0.006 0.011 -0.011 0.011 -0.000
(0.009) (0.010) (0.019) (0.009) (0.014)Industry Fixed Effects(Omitted: Agriculture, Mining, and Construction)Manufacturing 0.037*** 0.025* 0.020 0.028*
(0.013) (0.015) (0.024) (0.015)Transportation, Utilities & Warehousing 0.005 -0.006 -0.051* -0.008
(0.018) (0.020) (0.029) (0.019)Wholesale & Retail Trade 0.080*** 0.087*** 0.042** 0.082***
(0.013) (0.014) (0.021) (0.014)Services 0.091*** 0.096*** 0.083*** 0.086***
(0.014) (0.014) (0.022) (0.014)Accommodation, Food, and Entertainment 0.087*** 0.087*** 0.060* 0.082***
(0.015) (0.019) (0.035) (0.018)Other 0.146*** 0.144*** 0.143*** 0.138***
(0.016) (0.018) (0.022) (0.017)
Plant and firm controls Yes Yes Yes Yes Yes Yes2-digit NAICS FE No No No No No YesFiling year FE Yes Yes Yes No Yes Yes# of years searched FE Yes Yes Yes No Yes YesObservations 101,000 101,000 101,000 101000 101,000 101,000Adj. R-squared 0.0952 0.0872 0.0921 0.0295 0.0955 0.0984
46
Table 3First Stage
This table reports first stage results. The dependent variable is a dummy equal to one if a case is convertedfrom Chapter 11 reorganization to Chapter 7 liquidation. Column 1 reports results at the level of thebankruptcy filing, while Columns 2 and 3 report results at the level of the plant. In this and all otherregression tables, each observation is weighted by the inverse of the total number of plants belonging to thebankruptcy filing so as to give equal weight to each bankruptcy filing. The instrument we use is definedas the share of all other Chapter 11 cases that a judge converted to Chapter 7. The sample includes allfirms that filed for Chapter 11 bankruptcy between 1992 and 2005. Part of a group filing is an indicatorvariable equal to one if other related firms (e.g. subsidiaries of the same firm) also filed for bankruptcy at thesame time. Other controls are self-explanatory. All specifications contain 24 industry fixed effects and 2,361bankruptcy-division-by-year fixed effects. Standard errors, clustered at the division-by-year level, are shownin parentheses. *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively.
Dependent variable: Converted to Liquidation(1) (2) (3)
Share of other cases converted 0.581*** 0.581*** 0.580***(0.056) (0.054) (0.054)
Ln(employees at plant) 0.016***(0.003)
Plant age (years) -0.005***(0.000)
Ln(tot. employees at firm) -0.023*** -0.022*** -0.033***(0.003) (0.002) (0.004)
Ln(no. of plants at firm) -0.038*** -0.039*** -0.022***(0.006) (0.005) (0.006)
Part of a group filing -0.086*** -0.085*** -0.086***(0.011) (0.011) (0.011)
Unit of Observation Bankruptcy Plant Plant2-digit NAICS Fixed Effects Yes Yes YesDivision-year Fixed Effects Yes Yes YesObservations 28,000 129,000 129,000Adj. R-squared 0.102 0.165 0.170F-stat for instrument 107.2 114.9 113.5
47
Table 4First Stage Heterogeneity
This table reports first stage results for subsamples of the data by size and market characteris-tics. The first column shows the number of establishments in the group, and the second columnshows the number of firms. Column 3 shows the percentage of firms in the subsample that are con-verted to Chapter 7. Column 4 shows the coefficient on the instrument share converted when thefirst stage is run only on that subsample, with *** denoting statistical significance at the 1% level.These regressions include all control variables and fixed effects as in column 3 of Table 3. As dis-cussed in the text, these coefficients can also be interpreted as the share of firms that are sensitiveto the instrument. In column 5 we display the F -stat for the first stage. Finally, column 6 showsthe estimated percentage of firms in the group that would always be converted regardless of the judge.
Percent Coefficient on Fraction ofObservations # of firms Liquidated share converted F-stat always takers(1) (2) (3) (4) (5) (6)
Full Sample 129,000 28,000 41 0.581*** 114.10 18
Employees in firm:0-5 8,000 7,000 44 0.492*** 19.51 256-25 11,000 10,500 44 0.559*** 33.59 2126-100 11,000 6,500 41 0.880*** 66.87 6101-1000 22,000 3,000 29 0.521*** 8.74 9>1000 77,000 1,000 13 0.262 1.34 5
Market Characteristics:Thick markets 64,000 12,500 41 0.557*** 41.93 19Thin markets 65,000 17,000 40 0.603*** 77.11 16High access to capital 50,000 12,000 40 0.664*** 56.65 14Low access to capital 49,000 9,500 43 0.484*** 24.23 23
48
Table 5Random Judge Assignment
This table reports randomization tests to illustrate the random assignment of judges to bankruptcy filingswithin a division. The dependent variable is the share of Chapter 11 cases that a judge ever converted toChapter 7, which we use as an instrumental variable. All the regressions are at the plant level. Column1 contains only division-by-year fixed effects as controls and is included to demonstrate that the R2 is notaffected by the inclusion of any controls in Columns 2 - 8. Plant-level controls include employment at thelocation in the 3 years prior to bankruptcy, in addition to the standard controls as in Table 3. Heterogeneitymeasures are as defined in the text, and other independent variables are self-explanatory. The sampleincludes all firms that filed for Chapter 11 bankruptcy between 1992 and 2005. Standard errors, clusteredat the division-by-year level, are shown in parentheses. *, **, and *** denote statistical significance at the10%, 5%, and 1% level, respectively.
Dependent variable: Share converted(1) (2) (3) (4) (5) (6) (7) (8)
Plant- and firm-level controls:Ln(employees at plant)0 0.0002 -0.0001 -0.0001 -0.0001 -0.0001 -0.0001 -0.0001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)Ln(employees at plant)�1 0.0004 0.0004 0.0004 0.0004 0.0004 0.0004
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Ln(employees at plant)�2 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Ln(employees at plant)�3 0.0003 0.0003 0.0003 0.0003 0.0003 0.0003
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Plant age (years) -0.0000 -0.0001 -0.0001 -0.0001 -0.0001 -0.0001 -0.0001
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Ln(tot. employees at firm) 0.0009 0.0010 0.0010 0.0010 0.0009 0.0009 0.0010
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)Ln(no. plants at firm) -0.0012 -0.0012 -0.0012 -0.0012 -0.0012 -0.0012 -0.0012
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)Part of a group filing 0.0014 0.0014 0.0014 0.0014 0.0014 0.0014 0.0014
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)Dummy =1 if above median:Heterogeneity measures:Market Thickness 0.0001 0.0000 0.0000
(0.001) (0.001) (0.001)Share of small business loans 0.0007 0.0006 0.0007
(0.001) (0.001) (0.001)3-year employment growth in county 0.0017 0.0017 0.0016
(0.001) (0.001) (0.001)Other economic conditions:No. of plants in county -0.0006
(0.001)Payroll per employee in county 0.0012
(0.001)No. of plants in industry 0.0061
(0.004)Payroll per employee in industry 0.0006
(0.002)3-year employment growth in industry -0.0015
(0.001)
2-digit NAICS fixed effects No Yes Yes Yes Yes Yes Yes YesDivision-year fixed effects Yes Yes Yes Yes Yes Yes Yes YesF-stat for joint significance of industry FE 0.791 0.786 0.791 0.791 0.785 0.794 0.766Observations 129,000 129,000 129,000 129,000 129,000 129,000 129,000 129,000Adj. R-squared 0.777 0.777 0.777 0.777 0.777 0.778 0.778 0.778
49
Table 6Liquidation and Plant Outcomes
This table reports regression results showing the effect of liquidation on four plant outcomes 5 years afterthe bankruptcy filing. Continues is an indicator equal to 1 if the plant has at least one employee and is stillowned by the original bankrupt firm 5 years after the bankruptcy filing. Holding Time is the number of yearsafter bankruptcy (from 0 to 5) until a plant is not occupied by the original bankrupt firm. Occupied is anindicator equal to 1 if the plant has at least one employee regardless of the occupant. Average employmentis the mean number of employees at the plant over the five years after the bankruptcy filing (similar resultsfor payrolls are presented in the appendix). For all four dependent variables we display regular OLS and2SLS estimates. All specifications contain the full set of control variables in Column 3 of Table 3, includingdivision-by-year and industry fixed effects. Standard errors, clustered at the division-by-year level, are shownin parentheses. *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively.
Dependent variable: Continues Holding Time Occupied Ln(Avg. Employment)Model: OLS IV-2SLS OLS IV-2SLS OLS IV-2SLS OLS IV-2SLS
(1) (2) (3) (4) (5) (6) (7) (8)
Liquidation -0.300*** -0.324*** -1.773*** -1.657*** -0.156*** -0.174** -0.565*** -0.416*(0.005) (0.061) (0.024) (0.297) (0.007) (0.079) (0.019) (0.217)
Control Variables Yes Yes Yes Yes Yes Yes Yes YesDiv x Year FE Yes Yes Yes Yes Yes Yes Yes YesObservations 129,000 129,000 129,000 129,000 129,000 129,000 129,000 129,000Adjusted R-squared 0.230 0.152 0.288 0.211 0.130 0.039 0.295 0.214
50
Table 7Heterogeneous Effects on Utilization - Market Thickness
This table shows how the effects of liquidation vary depending on the thickness of the local asset market.Using our measure of market thickness (defined in the text), we divide the sample into plants in thickcounties (above-median) or thin counties (below-median). We then present, in columns 1, 2, 4, and 5,regression results similar to those in Table 6 separately for each sub-sample. In columns 3 and 6, we interactliquidation and the indicator for above-median market thickness and run the regression on the full sample todetermine if liquidation’s impact is significantly different in thick markets. For these interacted regressions,in the first stage we interact the instrument share converted and the above median indicator to generate twoinstruments for the two endogenous variables. Further, all control variables and fixed effects are interactedwith above-median market thickness. Panel A focuses on the two dependent variables that have to do withplant continuation: continues and holding time. Panel B displays the results with regards to two measuresof utilization: occupied and ln(average employment). All dependent variables are measured 5 years afterbankruptcy and are defined as in Table 6. All regressions are estimated by 2SLS and contain the full setof control variables in Column 3 of Table 3, including division-by-year and industry fixed effects. Standarderrors, clustered at the division-by-year level, are shown in parentheses. *, **, and *** denote statisticalsignificance at the 10%, 5%, and 1% level, respectively.
Panel A: Plant continuationDependent variable: Continues Holding TimeAbove or below median: Above Below Interacted Above Below Interacted
(1) (2) (3) (4) (5) (6)
Liquidation -0.337*** -0.321*** -0.321*** -1.449*** -1.836*** -1.836***(0.101) (0.076) (0.076) (0.492) (0.360) (0.360)
Liquidation * above median -0.016 0.387(0.127) (0.609)
Control Variables Yes Yes Yes Yes Yes YesDiv x Year FE Yes Yes Yes Yes Yes YesObservations 64,000 65,000 129,000 64,000 65,000 129,000
Panel B: Utilization
Dependent variable: Occupied Ln(Avg. Employment)Above or below median: Above Below Interacted Above Below Interacted
(1) (2) (3) (4) (5) (6)
Liquidation 0.080 -0.324*** -0.324*** 0.190 -0.790*** -0.790***(0.129) (0.109) (0.109) (0.413) (0.278) (0.278)
Liquidation * above median 0.404** 0.980**(0.169) (0.498)
Control Variables Yes Yes Yes Yes Yes YesDiv x Year FE Yes Yes Yes Yes Yes YesObservations 64,000 65,000 129,000 64,000 65,000 129,000
51
Table 8Heterogeneous Effects on Utilization - Access to Capital
This table shows how the effects of liquidation vary depending on the access to small business finance in thelocal market. Counties with an above-median share of bank loans going to small firms (firms with less than$1 million in annual revenue) are defined as having high access to capital, while below-median counties havelow access to capital. We then present, in columns 1, 2, 4, and 5, regression results similar to those in Table6 separately for plants in above- and below-median counties. In columns 3 and 6, we interact liquidationand the indicator for above-median access to capital and run the regression on the full sample to determineif liquidation’s impact is significantly different in markets with high access to capital. For these interactedregressions, in the first stage we interact the instrument share converted and the above median indicator togenerate two instruments for the two endogenous variables. Further, all control variables and fixed effectsare interacted with above-median access to capital . Panel A focuses on the two dependent variables thathave to do with plant continuation: continues and holding time. Panel B displays the results with regardsto two measures of utilization: occupied and ln(average employment). All dependent variables are measured5 years after bankruptcy and are defined as in Table 6. All regressions are estimated by 2SLS and containthe full set of control variables in Column 3 of Table 3, including division-by-year and industry fixed effects.Standard errors, clustered at the division-by-year level, are shown in parentheses. *, **, and *** denotestatistical significance at the 10%, 5%, and 1% level, respectively.
Panel A: Plant continuationDependent variable: Continues Holding TimeAbove or below median: Above Below Interacted Above Below Interacted
(1) (2) (3) (5) (6) (6)
Liquidation -0.273*** -0.341*** -0.341*** -1.402*** -1.999*** -1.999***(0.083) (0.126) (0.126) (0.413) (0.569) (0.569)
Liquidation * above median 0.069 0.596(0.151) (0.703)
Control Variables Yes Yes Yes Yes Yes YesDiv x Year FE Yes Yes Yes Yes Yes YesObservations 50,000 49,000 99,000 50,000 49,000 99,000
Panel B: Utilization
Dependent variable: Occupied Ln(Avg. Employment)Above or below median: Above Below Interacted Above Below Interacted
(3) (4) (3) (5) (6) (3)
Liquidation -0.018 -0.450** -0.450** -0.206 -1.197** -1.200**(0.111) (0.193) (0.193) (0.310) (0.480) (0.480)
Liquidation * above median 0.432* 0.992*(0.223) (0.571)
Control Variables Yes Yes Yes Yes Yes YesDiv x Year FE Yes Yes Yes Yes Yes YesObservations 50,000 49,000 99,000 50,000 49,000 99,000
52
Table 9Controlling for benchmark utilization
This table shows that liquidation affects plant utilization even when controlling for local utilization rates.Benchmark utilization rates come from a 5% random sample of establishments in the same county as eachbankrupt plant, with utilization measured 5 years after bankruptcy in an identical manner to the bankruptplant. Panel A focuses on liquidation’s impact on establishment occupancy, similar to column 6 of Table6, with the dependent variable being the occupied dummy for the bankrupt plant. High occupancy countyis a dummy indicating that a bankrupt plant is in a county with above-median benchmark occupancy, andliquidation*high occupancy is the interaction of this dummy with the liquidation indicator. Liquidation*thickmarket and liquidation*high access to capital are interactions between the liquidation indicator and localmarket conditions, as in Tables 7 and 8. For all interaction terms we also include the interaction betweenthe instrument and that variable in the first stage regression. Panel B is similar to Panel A, but insteadshowing the effect on ln(avg. employment). All regressions are estimated by 2SLS and contain the full setof control variables in Column 3 of Table 3, including division-by-year and industry fixed effects. Standarderrors, clustered at the division-by-year level, are shown in parentheses. *, **, and *** denote statisticalsignificance at the 10%, 5%, and 1% level, respectively.
Panel A: OccupiedDependent variable: Occupied
(1) (2) (3) (4) (5)
Liquidation -0.173** -0.321*** -0.295** -0.448** -0.445**(0.079) (0.109) (0.118) (0.193) (0.196)
High occupancy county 0.045*** 0.057*** 0.074*** 0.044*** 0.047(0.008) (0.010) (0.026) (0.013) (0.031)
Liquidation * Thick Market 0.399** 0.402**(0.169) (0.169)
Liquiation * High Access to Capital 0.433* 0.433*(0.223) (0.223)
Liquidation * High Occupancy -0.044 -0.006(0.062) (0.069)
Control Variables Yes Yes Yes Yes YesDiv x Year FE Yes Yes Yes Yes YesObservations 129,000 129,000 129,000 99,000 99,000
Panel B: Ln(Avg. Employment)Dependent variable: Ln(Avg. Employment)
(1) (2) (3) (4) (5)
Liquidation -0.408* -0.786*** -0.733** -1.193** -1.116**(0.217) (0.279) (0.306) (0.481) (0.494)
High occupancy county 0.088*** 0.100*** 0.135* 0.079** 0.133(0.020) (0.026) (0.072) (0.036) (0.089)
Liquidation * Thick Market 0.971* 0.977**(0.498) (0.497)
Liquiation * High Access to Capital 0.992* 0.994*(0.572) (0.572)
Liquidation * High Occupancy -0.089 -0.131(0.165) (0.193)
Control Variables Yes Yes Yes Yes YesDiv x Year FE Yes Yes Yes Yes YesObservations 129,000 129,000 129,000 99,000 99,000
53